blob: 7a5e6f9717f510471fb95c9440e6fee08e4ff626 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -080041#include <linux/kasan.h>
42
David Howells07f3f052006-09-30 20:52:18 +020043#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050044#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Nick Piggin789680d2011-01-07 17:49:30 +110046/*
47 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110048 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040049 * - i_dentry, d_u.d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110050 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100054 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110055 * - the dcache lru lists and counters
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110060 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110061 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110062 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
Al Viro946e51f2014-10-26 19:19:16 -040064 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110065 *
66 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110067 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110068 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100069 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110070 * dcache_hash_bucket lock
71 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110072 *
Nick Pigginda502952011-01-07 17:49:33 +110073 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
78 *
79 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110080 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
83 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080084int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
Al Viro74c3cbe2007-07-22 08:04:18 -040087__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Nick Piggin949854d2011-01-07 17:49:37 +110089EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Christoph Lametere18b8902006-12-06 20:33:20 -080091static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
97 *
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
100 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8387ff22016-06-10 07:51:30 -0700107static inline struct hlist_bl_head *d_hash(unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100108{
George Spelvin703b5fa2016-06-10 00:22:12 -0400109 return dentry_hashtable + (hash >> (32 - d_hash_shift));
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100110}
111
Al Viro94bdd652016-04-15 02:42:04 -0400112#define IN_LOOKUP_SHIFT 10
113static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
114
115static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
116 unsigned int hash)
117{
118 hash += (unsigned long) parent / L1_CACHE_BYTES;
119 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
120}
121
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/* Statistics gathering. */
124struct dentry_stat_t dentry_stat = {
125 .age_limit = 45,
126};
127
Glauber Costa3942c072013-08-28 10:17:53 +1000128static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000129static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400130
131#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000132
133/*
134 * Here we resort to our own counters instead of using generic per-cpu counters
135 * for consistency with what the vfs inode code does. We are expected to harvest
136 * better code and performance by having our own specialized counters.
137 *
138 * Please note that the loop is done over all possible CPUs, not over all online
139 * CPUs. The reason for this is that we don't want to play games with CPUs going
140 * on and off. If one of them goes off, we will just keep their counters.
141 *
142 * glommer: See cffbc8a for details, and if you ever intend to change this,
143 * please update all vfs counters to match.
144 */
Glauber Costa3942c072013-08-28 10:17:53 +1000145static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100146{
147 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000148 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry, i);
151 return sum < 0 ? 0 : sum;
152}
153
Dave Chinner62d36c72013-08-28 10:17:54 +1000154static long get_nr_dentry_unused(void)
155{
156 int i;
157 long sum = 0;
158 for_each_possible_cpu(i)
159 sum += per_cpu(nr_dentry_unused, i);
160 return sum < 0 ? 0 : sum;
161}
162
Joe Perches1f7e0612014-06-06 14:38:05 -0700163int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400164 size_t *lenp, loff_t *ppos)
165{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100166 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000167 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000168 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400169}
170#endif
171
Linus Torvalds5483f182012-03-04 15:51:42 -0800172/*
173 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
174 * The strings are both count bytes long, and count is non-zero.
175 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700176#ifdef CONFIG_DCACHE_WORD_ACCESS
177
178#include <asm/word-at-a-time.h>
179/*
180 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
181 * aligned allocation for this particular component. We don't
182 * strictly need the load_unaligned_zeropad() safety, but it
183 * doesn't hurt either.
184 *
185 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
186 * need the careful unaligned handling.
187 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700188static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800189{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800190 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800191
192 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700193 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700194 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 if (tcount < sizeof(unsigned long))
196 break;
197 if (unlikely(a != b))
198 return 1;
199 cs += sizeof(unsigned long);
200 ct += sizeof(unsigned long);
201 tcount -= sizeof(unsigned long);
202 if (!tcount)
203 return 0;
204 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000205 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800206 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700207}
208
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800209#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700210
Linus Torvalds94753db52012-05-10 12:19:19 -0700211static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212{
Linus Torvalds5483f182012-03-04 15:51:42 -0800213 do {
214 if (*cs != *ct)
215 return 1;
216 cs++;
217 ct++;
218 tcount--;
219 } while (tcount);
220 return 0;
221}
222
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700223#endif
224
Linus Torvalds94753db52012-05-10 12:19:19 -0700225static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
226{
Linus Torvalds94753db52012-05-10 12:19:19 -0700227 /*
228 * Be careful about RCU walk racing with rename:
He Kuangae0a8432016-03-26 09:12:10 +0000229 * use 'lockless_dereference' to fetch the name pointer.
Linus Torvalds94753db52012-05-10 12:19:19 -0700230 *
231 * NOTE! Even if a rename will mean that the length
232 * was not loaded atomically, we don't care. The
233 * RCU walk will check the sequence count eventually,
234 * and catch it. And we won't overrun the buffer,
235 * because we're reading the name pointer atomically,
236 * and a dentry name is guaranteed to be properly
237 * terminated with a NUL byte.
238 *
239 * End result: even if 'len' is wrong, we'll exit
240 * early because the data cannot match (there can
241 * be no NUL in the ct/tcount data)
242 */
He Kuangae0a8432016-03-26 09:12:10 +0000243 const unsigned char *cs = lockless_dereference(dentry->d_name.name);
244
Linus Torvalds6326c712012-05-21 16:14:04 -0700245 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700246}
247
Al Viro8d85b482014-09-29 14:54:27 -0400248struct external_name {
249 union {
250 atomic_t count;
251 struct rcu_head head;
252 } u;
253 unsigned char name[];
254};
255
256static inline struct external_name *external_name(struct dentry *dentry)
257{
258 return container_of(dentry->d_name.name, struct external_name, name[0]);
259}
260
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400261static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400263 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
264
Al Viro8d85b482014-09-29 14:54:27 -0400265 kmem_cache_free(dentry_cache, dentry);
266}
267
268static void __d_free_external(struct rcu_head *head)
269{
270 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400271 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 kmem_cache_free(dentry_cache, dentry);
273}
274
Al Viro810bb172014-10-12 12:45:37 -0400275static inline int dname_external(const struct dentry *dentry)
276{
277 return dentry->d_name.name != dentry->d_iname;
278}
279
Al Viroad25f112017-07-07 14:51:19 -0400280void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
281{
282 spin_lock(&dentry->d_lock);
283 if (unlikely(dname_external(dentry))) {
284 struct external_name *p = external_name(dentry);
285 atomic_inc(&p->u.count);
286 spin_unlock(&dentry->d_lock);
287 name->name = p->name;
288 } else {
289 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
290 spin_unlock(&dentry->d_lock);
291 name->name = name->inline_name;
292 }
293}
294EXPORT_SYMBOL(take_dentry_name_snapshot);
295
296void release_dentry_name_snapshot(struct name_snapshot *name)
297{
298 if (unlikely(name->name != name->inline_name)) {
299 struct external_name *p;
300 p = container_of(name->name, struct external_name, name[0]);
301 if (unlikely(atomic_dec_and_test(&p->u.count)))
302 kfree_rcu(p, u.head);
303 }
304}
305EXPORT_SYMBOL(release_dentry_name_snapshot);
306
David Howells4bf46a22015-03-05 14:09:22 +0000307static inline void __d_set_inode_and_type(struct dentry *dentry,
308 struct inode *inode,
309 unsigned type_flags)
310{
311 unsigned flags;
312
313 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000314 flags = READ_ONCE(dentry->d_flags);
315 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
316 flags |= type_flags;
317 WRITE_ONCE(dentry->d_flags, flags);
318}
319
David Howells4bf46a22015-03-05 14:09:22 +0000320static inline void __d_clear_type_and_inode(struct dentry *dentry)
321{
322 unsigned flags = READ_ONCE(dentry->d_flags);
323
324 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
325 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000326 dentry->d_inode = NULL;
327}
328
Al Virob4f03542014-04-29 23:40:14 -0400329static void dentry_free(struct dentry *dentry)
330{
Al Viro946e51f2014-10-26 19:19:16 -0400331 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400332 if (unlikely(dname_external(dentry))) {
333 struct external_name *p = external_name(dentry);
334 if (likely(atomic_dec_and_test(&p->u.count))) {
335 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
336 return;
337 }
338 }
Al Virob4f03542014-04-29 23:40:14 -0400339 /* if dentry was never visible to RCU, immediate free is OK */
340 if (!(dentry->d_flags & DCACHE_RCUACCESS))
341 __d_free(&dentry->d_u.d_rcu);
342 else
343 call_rcu(&dentry->d_u.d_rcu, __d_free);
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/*
347 * Release the dentry's inode, using the filesystem
Al Viro550dce02016-05-29 20:13:30 -0400348 * d_iput() operation if defined.
Nick Piggin31e6b012011-01-07 17:49:52 +1100349 */
350static void dentry_unlink_inode(struct dentry * dentry)
351 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100352 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100353{
354 struct inode *inode = dentry->d_inode;
Al Viro550dce02016-05-29 20:13:30 -0400355 bool hashed = !d_unhashed(dentry);
Al Viroa528aca2016-02-29 12:12:46 -0500356
Al Viro550dce02016-05-29 20:13:30 -0400357 if (hashed)
358 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000359 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400360 hlist_del_init(&dentry->d_u.d_alias);
Al Viro550dce02016-05-29 20:13:30 -0400361 if (hashed)
362 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100363 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100364 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100365 if (!inode->i_nlink)
366 fsnotify_inoderemove(inode);
367 if (dentry->d_op && dentry->d_op->d_iput)
368 dentry->d_op->d_iput(dentry, inode);
369 else
370 iput(inode);
371}
372
373/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400374 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
375 * is in use - which includes both the "real" per-superblock
376 * LRU list _and_ the DCACHE_SHRINK_LIST use.
377 *
378 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
379 * on the shrink list (ie not on the superblock LRU list).
380 *
381 * The per-cpu "nr_dentry_unused" counters are updated with
382 * the DCACHE_LRU_LIST bit.
383 *
384 * These helper functions make sure we always follow the
385 * rules. d_lock must be held by the caller.
386 */
387#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
388static void d_lru_add(struct dentry *dentry)
389{
390 D_FLAG_VERIFY(dentry, 0);
391 dentry->d_flags |= DCACHE_LRU_LIST;
392 this_cpu_inc(nr_dentry_unused);
393 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
394}
395
396static void d_lru_del(struct dentry *dentry)
397{
398 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
399 dentry->d_flags &= ~DCACHE_LRU_LIST;
400 this_cpu_dec(nr_dentry_unused);
401 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
402}
403
404static void d_shrink_del(struct dentry *dentry)
405{
406 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
407 list_del_init(&dentry->d_lru);
408 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
409 this_cpu_dec(nr_dentry_unused);
410}
411
412static void d_shrink_add(struct dentry *dentry, struct list_head *list)
413{
414 D_FLAG_VERIFY(dentry, 0);
415 list_add(&dentry->d_lru, list);
416 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
417 this_cpu_inc(nr_dentry_unused);
418}
419
420/*
421 * These can only be called under the global LRU lock, ie during the
422 * callback for freeing the LRU list. "isolate" removes it from the
423 * LRU lists entirely, while shrink_move moves it to the indicated
424 * private list.
425 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800426static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400427{
428 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
429 dentry->d_flags &= ~DCACHE_LRU_LIST;
430 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800431 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400432}
433
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800434static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
435 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400436{
437 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
438 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800439 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400440}
441
442/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000443 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700444 */
445static void dentry_lru_add(struct dentry *dentry)
446{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400447 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
448 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700449}
450
Miklos Szeredid52b9082007-05-08 00:23:46 -0700451/**
Nick Piggin789680d2011-01-07 17:49:30 +1100452 * d_drop - drop a dentry
453 * @dentry: dentry to drop
454 *
455 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
456 * be found through a VFS lookup any more. Note that this is different from
457 * deleting the dentry - d_delete will try to mark the dentry negative if
458 * possible, giving a successful _negative_ lookup, while d_drop will
459 * just make the cache lookup fail.
460 *
461 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
462 * reason (NFS timeouts or autofs deletes).
463 *
NeilBrown3aa66ba2017-11-10 15:45:41 +1100464 * __d_drop requires dentry->d_lock
465 * ___d_drop doesn't mark dentry as "unhashed"
466 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
Nick Piggin789680d2011-01-07 17:49:30 +1100467 */
NeilBrown3aa66ba2017-11-10 15:45:41 +1100468static void ___d_drop(struct dentry *dentry)
Nick Piggin789680d2011-01-07 17:49:30 +1100469{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700470 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400471 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400472 /*
473 * Hashed dentries are normally on the dentry hashtable,
474 * with the exception of those newly allocated by
475 * d_obtain_alias, which are always IS_ROOT:
476 */
477 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400478 b = &dentry->d_sb->s_anon;
479 else
Linus Torvalds8387ff22016-06-10 07:51:30 -0700480 b = d_hash(dentry->d_name.hash);
Al Virob61625d2013-10-04 11:09:01 -0400481
482 hlist_bl_lock(b);
483 __hlist_bl_del(&dentry->d_hash);
Al Virob61625d2013-10-04 11:09:01 -0400484 hlist_bl_unlock(b);
Al Virod6141462016-07-28 13:05:50 -0400485 /* After this call, in-progress rcu-walk path lookup will fail. */
486 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin789680d2011-01-07 17:49:30 +1100487 }
488}
NeilBrown3aa66ba2017-11-10 15:45:41 +1100489
490void __d_drop(struct dentry *dentry)
491{
492 ___d_drop(dentry);
493 dentry->d_hash.pprev = NULL;
494}
Nick Piggin789680d2011-01-07 17:49:30 +1100495EXPORT_SYMBOL(__d_drop);
496
497void d_drop(struct dentry *dentry)
498{
Nick Piggin789680d2011-01-07 17:49:30 +1100499 spin_lock(&dentry->d_lock);
500 __d_drop(dentry);
501 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100502}
503EXPORT_SYMBOL(d_drop);
504
Al Viroba65dc52016-06-10 11:32:47 -0400505static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
506{
507 struct dentry *next;
508 /*
509 * Inform d_walk() and shrink_dentry_list() that we are no longer
510 * attached to the dentry tree
511 */
512 dentry->d_flags |= DCACHE_DENTRY_KILLED;
513 if (unlikely(list_empty(&dentry->d_child)))
514 return;
515 __list_del_entry(&dentry->d_child);
516 /*
517 * Cursors can move around the list of children. While we'd been
518 * a normal list member, it didn't matter - ->d_child.next would've
519 * been updated. However, from now on it won't be and for the
520 * things like d_walk() it might end up with a nasty surprise.
521 * Normally d_walk() doesn't care about cursors moving around -
522 * ->d_lock on parent prevents that and since a cursor has no children
523 * of its own, we get through it without ever unlocking the parent.
524 * There is one exception, though - if we ascend from a child that
525 * gets killed as soon as we unlock it, the next sibling is found
526 * using the value left in its ->d_child.next. And if _that_
527 * pointed to a cursor, and cursor got moved (e.g. by lseek())
528 * before d_walk() regains parent->d_lock, we'll end up skipping
529 * everything the cursor had been moved past.
530 *
531 * Solution: make sure that the pointer left behind in ->d_child.next
532 * points to something that won't be moving around. I.e. skip the
533 * cursors.
534 */
535 while (dentry->d_child.next != &parent->d_subdirs) {
536 next = list_entry(dentry->d_child.next, struct dentry, d_child);
537 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
538 break;
539 dentry->d_child.next = next->d_child.next;
540 }
541}
542
Al Viroe55fd012014-05-28 13:51:12 -0400543static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100544{
Al Viro41edf272014-05-01 10:30:00 -0400545 struct dentry *parent = NULL;
546 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400547 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100548 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100549
Linus Torvalds0d984392013-09-08 13:46:52 -0700550 /*
551 * The dentry is now unrecoverably dead to the world.
552 */
553 lockref_mark_dead(&dentry->d_lockref);
554
Sage Weilf0023bc2011-10-28 10:02:42 -0700555 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700556 * inform the fs via d_prune that this dentry is about to be
557 * unhashed and destroyed.
558 */
Al Viro29266202014-05-30 11:39:02 -0400559 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800560 dentry->d_op->d_prune(dentry);
561
Al Viro01b60352014-04-29 23:42:52 -0400562 if (dentry->d_flags & DCACHE_LRU_LIST) {
563 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
564 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400565 }
Nick Piggin77812a12011-01-07 17:49:48 +1100566 /* if it was on the hash then remove it */
567 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400568 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400569 if (parent)
570 spin_unlock(&parent->d_lock);
Al Viro550dce02016-05-29 20:13:30 -0400571 if (dentry->d_inode)
572 dentry_unlink_inode(dentry);
573 else
574 spin_unlock(&dentry->d_lock);
Al Viro03b3b882014-04-29 15:45:28 -0400575 this_cpu_dec(nr_dentry);
576 if (dentry->d_op && dentry->d_op->d_release)
577 dentry->d_op->d_release(dentry);
578
Al Viro41edf272014-05-01 10:30:00 -0400579 spin_lock(&dentry->d_lock);
580 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
581 dentry->d_flags |= DCACHE_MAY_FREE;
582 can_free = false;
583 }
584 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400585 if (likely(can_free))
586 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400587}
588
589/*
590 * Finish off a dentry we've decided to kill.
591 * dentry->d_lock must be held, returns with it unlocked.
592 * If ref is non-zero, then decrement the refcount too.
593 * Returns dentry requiring refcount drop, or NULL if we're done.
594 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400595static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400596 __releases(dentry->d_lock)
597{
598 struct inode *inode = dentry->d_inode;
599 struct dentry *parent = NULL;
600
601 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
602 goto failed;
603
604 if (!IS_ROOT(dentry)) {
605 parent = dentry->d_parent;
606 if (unlikely(!spin_trylock(&parent->d_lock))) {
607 if (inode)
608 spin_unlock(&inode->i_lock);
609 goto failed;
610 }
611 }
612
613 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400614 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400615
616failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400617 spin_unlock(&dentry->d_lock);
Al Viroe55fd012014-05-28 13:51:12 -0400618 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100619}
620
Al Viro046b9612014-05-29 08:54:52 -0400621static inline struct dentry *lock_parent(struct dentry *dentry)
622{
623 struct dentry *parent = dentry->d_parent;
624 if (IS_ROOT(dentry))
625 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800626 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400627 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400628 if (likely(spin_trylock(&parent->d_lock)))
629 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400630 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400631 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400632again:
633 parent = ACCESS_ONCE(dentry->d_parent);
634 spin_lock(&parent->d_lock);
635 /*
636 * We can't blindly lock dentry until we are sure
637 * that we won't violate the locking order.
638 * Any changes of dentry->d_parent must have
639 * been done with parent->d_lock held, so
640 * spin_lock() above is enough of a barrier
641 * for checking if it's still our child.
642 */
643 if (unlikely(parent != dentry->d_parent)) {
644 spin_unlock(&parent->d_lock);
645 goto again;
646 }
Al Viro05f16fe2018-02-23 20:47:17 -0500647 if (parent != dentry) {
Linus Torvalds9f126002014-05-31 09:13:21 -0700648 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro05f16fe2018-02-23 20:47:17 -0500649 if (unlikely(dentry->d_lockref.count < 0)) {
650 spin_unlock(&parent->d_lock);
651 parent = NULL;
652 }
653 } else {
Al Viro046b9612014-05-29 08:54:52 -0400654 parent = NULL;
Al Viro05f16fe2018-02-23 20:47:17 -0500655 }
656 rcu_read_unlock();
Al Viro046b9612014-05-29 08:54:52 -0400657 return parent;
658}
659
Linus Torvalds360f5472015-01-09 15:19:03 -0800660/*
661 * Try to do a lockless dput(), and return whether that was successful.
662 *
663 * If unsuccessful, we return false, having already taken the dentry lock.
664 *
665 * The caller needs to hold the RCU read lock, so that the dentry is
666 * guaranteed to stay around even if the refcount goes down to zero!
667 */
668static inline bool fast_dput(struct dentry *dentry)
669{
670 int ret;
671 unsigned int d_flags;
672
673 /*
674 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100675 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800676 */
677 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
678 return lockref_put_or_lock(&dentry->d_lockref);
679
680 /*
681 * .. otherwise, we can try to just decrement the
682 * lockref optimistically.
683 */
684 ret = lockref_put_return(&dentry->d_lockref);
685
686 /*
687 * If the lockref_put_return() failed due to the lock being held
688 * by somebody else, the fast path has failed. We will need to
689 * get the lock, and then check the count again.
690 */
691 if (unlikely(ret < 0)) {
692 spin_lock(&dentry->d_lock);
693 if (dentry->d_lockref.count > 1) {
694 dentry->d_lockref.count--;
695 spin_unlock(&dentry->d_lock);
696 return 1;
697 }
698 return 0;
699 }
700
701 /*
702 * If we weren't the last ref, we're done.
703 */
704 if (ret)
705 return 1;
706
707 /*
708 * Careful, careful. The reference count went down
709 * to zero, but we don't hold the dentry lock, so
710 * somebody else could get it again, and do another
711 * dput(), and we need to not race with that.
712 *
713 * However, there is a very special and common case
714 * where we don't care, because there is nothing to
715 * do: the dentry is still hashed, it does not have
716 * a 'delete' op, and it's referenced and already on
717 * the LRU list.
718 *
719 * NOTE! Since we aren't locked, these values are
720 * not "stable". However, it is sufficient that at
721 * some point after we dropped the reference the
722 * dentry was hashed and the flags had the proper
723 * value. Other dentry users may have re-gotten
724 * a reference to the dentry and change that, but
725 * our work is done - we can leave the dentry
726 * around with a zero refcount.
727 */
728 smp_rmb();
729 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100730 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800731
732 /* Nothing to do? Dropping the reference was all we needed? */
733 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
734 return 1;
735
736 /*
737 * Not the fast normal case? Get the lock. We've already decremented
738 * the refcount, but we'll need to re-check the situation after
739 * getting the lock.
740 */
741 spin_lock(&dentry->d_lock);
742
743 /*
744 * Did somebody else grab a reference to it in the meantime, and
745 * we're no longer the last user after all? Alternatively, somebody
746 * else could have killed it and marked it dead. Either way, we
747 * don't need to do anything else.
748 */
749 if (dentry->d_lockref.count) {
750 spin_unlock(&dentry->d_lock);
751 return 1;
752 }
753
754 /*
755 * Re-get the reference we optimistically dropped. We hold the
756 * lock, and we just tested that it was zero, so we can just
757 * set it to 1.
758 */
759 dentry->d_lockref.count = 1;
760 return 0;
761}
762
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764/*
765 * This is dput
766 *
767 * This is complicated by the fact that we do not want to put
768 * dentries that are no longer on any hash chain on the unused
769 * list: we'd much rather just get rid of them immediately.
770 *
771 * However, that implies that we have to traverse the dentry
772 * tree upwards to the parents which might _also_ now be
773 * scheduled for deletion (it may have been only waiting for
774 * its last child to go away).
775 *
776 * This tail recursion is done by hand as we don't want to depend
777 * on the compiler to always get this right (gcc generally doesn't).
778 * Real recursion would eat up our stack space.
779 */
780
781/*
782 * dput - release a dentry
783 * @dentry: dentry to release
784 *
785 * Release a dentry. This will drop the usage count and if appropriate
786 * call the dentry unlink method as well as removing it from the queues and
787 * releasing its resources. If the parent dentries were scheduled for release
788 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790void dput(struct dentry *dentry)
791{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700792 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return;
794
795repeat:
Wei Fang47be6182016-07-06 11:32:20 +0800796 might_sleep();
797
Linus Torvalds360f5472015-01-09 15:19:03 -0800798 rcu_read_lock();
799 if (likely(fast_dput(dentry))) {
800 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800802 }
803
804 /* Slow case: now with the dentry lock held */
805 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Al Viro85c7f812016-04-14 19:52:13 -0400807 WARN_ON(d_in_lookup(dentry));
808
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700809 /* Unreachable? Get rid of it */
810 if (unlikely(d_unhashed(dentry)))
811 goto kill_it;
812
Al Viro75a6f822015-07-08 02:42:38 +0100813 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
814 goto kill_it;
815
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700816 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100818 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 }
Nick Piggin265ac902010-10-10 05:36:24 -0400820
Linus Torvalds358eec12013-10-31 15:43:02 -0700821 if (!(dentry->d_flags & DCACHE_REFERENCED))
822 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400823 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400824
Waiman Long98474232013-08-28 18:24:59 -0700825 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100826 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return;
828
Miklos Szeredid52b9082007-05-08 00:23:46 -0700829kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400830 dentry = dentry_kill(dentry);
Wei Fang47be6182016-07-06 11:32:20 +0800831 if (dentry) {
832 cond_resched();
Miklos Szeredid52b9082007-05-08 00:23:46 -0700833 goto repeat;
Wei Fang47be6182016-07-06 11:32:20 +0800834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700836EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100839/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100840static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841{
Waiman Long98474232013-08-28 18:24:59 -0700842 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
Nick Piggindc0474b2011-01-07 17:49:43 +1100845static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100846{
Waiman Long98474232013-08-28 18:24:59 -0700847 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100848}
849
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100850struct dentry *dget_parent(struct dentry *dentry)
851{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700852 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100853 struct dentry *ret;
854
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700855 /*
856 * Do optimistic parent lookup without any
857 * locking.
858 */
859 rcu_read_lock();
860 ret = ACCESS_ONCE(dentry->d_parent);
861 gotref = lockref_get_not_zero(&ret->d_lockref);
862 rcu_read_unlock();
863 if (likely(gotref)) {
864 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
865 return ret;
866 dput(ret);
867 }
868
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100869repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100870 /*
871 * Don't need rcu_dereference because we re-check it was correct under
872 * the lock.
873 */
874 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100875 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100876 spin_lock(&ret->d_lock);
877 if (unlikely(ret != dentry->d_parent)) {
878 spin_unlock(&ret->d_lock);
879 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100880 goto repeat;
881 }
Nick Piggina734eb42011-01-07 17:49:44 +1100882 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700883 BUG_ON(!ret->d_lockref.count);
884 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100885 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100886 return ret;
887}
888EXPORT_SYMBOL(dget_parent);
889
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890/**
891 * d_find_alias - grab a hashed alias of inode
892 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 *
894 * If inode has a hashed alias, or is a directory and has any alias,
895 * acquire the reference to alias and return it. Otherwise return NULL.
896 * Notice that if inode is a directory there can be only one alias and
897 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800898 * of a filesystem, or if the directory was renamed and d_revalidate
899 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700901 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500902 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500904static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
Nick Pigginda502952011-01-07 17:49:33 +1100906 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Nick Pigginda502952011-01-07 17:49:33 +1100908again:
909 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400910 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100911 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700913 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100914 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500916 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100917 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100918 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 return alias;
920 }
921 }
Nick Pigginda502952011-01-07 17:49:33 +1100922 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 }
Nick Pigginda502952011-01-07 17:49:33 +1100924 if (discon_alias) {
925 alias = discon_alias;
926 spin_lock(&alias->d_lock);
927 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500928 __dget_dlock(alias);
929 spin_unlock(&alias->d_lock);
930 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100931 }
932 spin_unlock(&alias->d_lock);
933 goto again;
934 }
935 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
937
Nick Pigginda502952011-01-07 17:49:33 +1100938struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
David Howells214fda12006-03-25 03:06:36 -0800940 struct dentry *de = NULL;
941
Al Virob3d9b7a2012-06-09 13:51:19 -0400942 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100943 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500944 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100945 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 return de;
948}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700949EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
951/*
952 * Try to kill dentries associated with this inode.
953 * WARNING: you must own a reference to inode.
954 */
955void d_prune_aliases(struct inode *inode)
956{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700957 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100959 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400960 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700962 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400963 struct dentry *parent = lock_parent(dentry);
964 if (likely(!dentry->d_lockref.count)) {
965 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800966 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400967 goto restart;
968 }
969 if (parent)
970 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 }
972 spin_unlock(&dentry->d_lock);
973 }
Nick Piggin873feea2011-01-07 17:50:06 +1100974 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700976EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400978static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Al Viro5c47e6d2014-04-29 16:13:18 -0400980 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700981
Miklos Szeredi60942f22014-05-02 15:38:39 -0400982 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400983 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400984 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100985 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400986 parent = lock_parent(dentry);
987
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000989 * The dispose list is isolated and dentries are not accounted
990 * to the LRU here, so we can simply remove it from the list
991 * here regardless of whether it is referenced or not.
992 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400993 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000994
995 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000997 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800999 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001000 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001001 if (parent)
1002 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 continue;
1004 }
Nick Piggin77812a12011-01-07 17:49:48 +11001005
Al Viro64fd72e2014-05-28 09:48:44 -04001006
1007 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1008 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1009 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001010 if (parent)
1011 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -04001012 if (can_free)
1013 dentry_free(dentry);
1014 continue;
1015 }
1016
Al Viroff2fde92014-05-28 13:59:13 -04001017 inode = dentry->d_inode;
1018 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001019 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001020 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001021 if (parent)
1022 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -04001023 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001024 }
Al Viroff2fde92014-05-28 13:59:13 -04001025
Al Viroff2fde92014-05-28 13:59:13 -04001026 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -04001027
Al Viro5c47e6d2014-04-29 16:13:18 -04001028 /*
1029 * We need to prune ancestors too. This is necessary to prevent
1030 * quadratic behavior of shrink_dcache_parent(), but is also
1031 * expected to be beneficial in reducing dentry cache
1032 * fragmentation.
1033 */
1034 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -04001035 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1036 parent = lock_parent(dentry);
1037 if (dentry->d_lockref.count != 1) {
1038 dentry->d_lockref.count--;
1039 spin_unlock(&dentry->d_lock);
1040 if (parent)
1041 spin_unlock(&parent->d_lock);
1042 break;
1043 }
1044 inode = dentry->d_inode; /* can't be NULL */
1045 if (unlikely(!spin_trylock(&inode->i_lock))) {
1046 spin_unlock(&dentry->d_lock);
1047 if (parent)
1048 spin_unlock(&parent->d_lock);
1049 cpu_relax();
1050 continue;
1051 }
1052 __dentry_kill(dentry);
1053 dentry = parent;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001056}
1057
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001058static enum lru_status dentry_lru_isolate(struct list_head *item,
1059 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001060{
1061 struct list_head *freeable = arg;
1062 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1063
1064
1065 /*
1066 * we are inverting the lru lock/dentry->d_lock here,
1067 * so use a trylock. If we fail to get the lock, just skip
1068 * it
1069 */
1070 if (!spin_trylock(&dentry->d_lock))
1071 return LRU_SKIP;
1072
1073 /*
1074 * Referenced dentries are still in use. If they have active
1075 * counts, just remove them from the LRU. Otherwise give them
1076 * another pass through the LRU.
1077 */
1078 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001079 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001080 spin_unlock(&dentry->d_lock);
1081 return LRU_REMOVED;
1082 }
1083
1084 if (dentry->d_flags & DCACHE_REFERENCED) {
1085 dentry->d_flags &= ~DCACHE_REFERENCED;
1086 spin_unlock(&dentry->d_lock);
1087
1088 /*
1089 * The list move itself will be made by the common LRU code. At
1090 * this point, we've dropped the dentry->d_lock but keep the
1091 * lru lock. This is safe to do, since every list movement is
1092 * protected by the lru lock even if both locks are held.
1093 *
1094 * This is guaranteed by the fact that all LRU management
1095 * functions are intermediated by the LRU API calls like
1096 * list_lru_add and list_lru_del. List movement in this file
1097 * only ever occur through this functions or through callbacks
1098 * like this one, that are called from the LRU API.
1099 *
1100 * The only exceptions to this are functions like
1101 * shrink_dentry_list, and code that first checks for the
1102 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1103 * operating only with stack provided lists after they are
1104 * properly isolated from the main list. It is thus, always a
1105 * local access.
1106 */
1107 return LRU_ROTATE;
1108 }
1109
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001110 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001111 spin_unlock(&dentry->d_lock);
1112
1113 return LRU_REMOVED;
1114}
1115
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001116/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001117 * prune_dcache_sb - shrink the dcache
1118 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001119 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001120 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001121 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1122 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001123 * function.
1124 *
1125 * This function may fail to free any resources if all the dentries are in
1126 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001127 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001128long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001129{
Dave Chinnerf6041562013-08-28 10:18:00 +10001130 LIST_HEAD(dispose);
1131 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001132
Vladimir Davydov503c3582015-02-12 14:58:47 -08001133 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1134 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001135 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001136 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
Glauber Costa4e717f52013-08-28 10:18:03 +10001139static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001140 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001141{
Glauber Costa4e717f52013-08-28 10:18:03 +10001142 struct list_head *freeable = arg;
1143 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001144
Glauber Costa4e717f52013-08-28 10:18:03 +10001145 /*
1146 * we are inverting the lru lock/dentry->d_lock here,
1147 * so use a trylock. If we fail to get the lock, just skip
1148 * it
1149 */
1150 if (!spin_trylock(&dentry->d_lock))
1151 return LRU_SKIP;
1152
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001153 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001154 spin_unlock(&dentry->d_lock);
1155
1156 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001157}
1158
Glauber Costa4e717f52013-08-28 10:18:03 +10001159
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001160/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 * shrink_dcache_sb - shrink dcache for a superblock
1162 * @sb: superblock
1163 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001164 * Shrink the dcache for the specified super block. This is used to free
1165 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001167void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
Glauber Costa4e717f52013-08-28 10:18:03 +10001169 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001170
Glauber Costa4e717f52013-08-28 10:18:03 +10001171 do {
1172 LIST_HEAD(dispose);
1173
1174 freed = list_lru_walk(&sb->s_dentry_lru,
Sahitya Tummalaa9aa6522a2017-07-10 15:50:00 -07001175 dentry_lru_isolate_shrink, &dispose, 1024);
Glauber Costa4e717f52013-08-28 10:18:03 +10001176
1177 this_cpu_sub(nr_dentry_unused, freed);
1178 shrink_dentry_list(&dispose);
Sahitya Tummalaa9aa6522a2017-07-10 15:50:00 -07001179 cond_resched();
1180 } while (list_lru_count(&sb->s_dentry_lru) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001182EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001185 * enum d_walk_ret - action to talke during tree walk
1186 * @D_WALK_CONTINUE: contrinue walk
1187 * @D_WALK_QUIT: quit walk
1188 * @D_WALK_NORETRY: quit when retry is needed
1189 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001191enum d_walk_ret {
1192 D_WALK_CONTINUE,
1193 D_WALK_QUIT,
1194 D_WALK_NORETRY,
1195 D_WALK_SKIP,
1196};
1197
1198/**
1199 * d_walk - walk the dentry tree
1200 * @parent: start of walk
1201 * @data: data passed to @enter() and @finish()
1202 * @enter: callback when first entering the dentry
1203 * @finish: callback when successfully finished the walk
1204 *
1205 * The @enter() and @finish() callbacks are called with d_lock held.
1206 */
1207static void d_walk(struct dentry *parent, void *data,
1208 enum d_walk_ret (*enter)(void *, struct dentry *),
1209 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210{
Nick Piggin949854d2011-01-07 17:49:37 +11001211 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001213 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001214 enum d_walk_ret ret;
1215 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001216
Nick Piggin58db63d2011-01-07 17:49:39 +11001217again:
Al Viro48f5ec22013-09-09 15:22:25 -04001218 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001219 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001220 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001221
1222 ret = enter(data, this_parent);
1223 switch (ret) {
1224 case D_WALK_CONTINUE:
1225 break;
1226 case D_WALK_QUIT:
1227 case D_WALK_SKIP:
1228 goto out_unlock;
1229 case D_WALK_NORETRY:
1230 retry = false;
1231 break;
1232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233repeat:
1234 next = this_parent->d_subdirs.next;
1235resume:
1236 while (next != &this_parent->d_subdirs) {
1237 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001238 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001240
Al Viroba65dc52016-06-10 11:32:47 -04001241 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1242 continue;
1243
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001244 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001245
1246 ret = enter(data, dentry);
1247 switch (ret) {
1248 case D_WALK_CONTINUE:
1249 break;
1250 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001251 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001252 goto out_unlock;
1253 case D_WALK_NORETRY:
1254 retry = false;
1255 break;
1256 case D_WALK_SKIP:
1257 spin_unlock(&dentry->d_lock);
1258 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001259 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001262 spin_unlock(&this_parent->d_lock);
1263 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001265 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 goto repeat;
1267 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001268 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 }
1270 /*
1271 * All done at this level ... ascend and resume the search.
1272 */
Al Viroca5358e2014-10-26 19:31:10 -04001273 rcu_read_lock();
1274ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001276 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001277 this_parent = child->d_parent;
1278
Al Viro31dec132013-10-25 17:04:27 -04001279 spin_unlock(&child->d_lock);
1280 spin_lock(&this_parent->d_lock);
1281
Al Viroca5358e2014-10-26 19:31:10 -04001282 /* might go back up the wrong parent if we have had a rename. */
1283 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001284 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001285 /* go into the first sibling still alive */
1286 do {
1287 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001288 if (next == &this_parent->d_subdirs)
1289 goto ascend;
1290 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001291 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001292 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 goto resume;
1294 }
Al Viroca5358e2014-10-26 19:31:10 -04001295 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001296 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001297 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001298 if (finish)
1299 finish(data);
1300
1301out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001302 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001303 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001304 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001305
1306rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001307 spin_unlock(&this_parent->d_lock);
1308 rcu_read_unlock();
1309 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001310 if (!retry)
1311 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001312 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001313 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001315
1316/*
1317 * Search for at least 1 mount point in the dentry's subdirs.
1318 * We descend to the next level whenever the d_subdirs
1319 * list is non-empty and continue searching.
1320 */
1321
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001322static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1323{
1324 int *ret = data;
1325 if (d_mountpoint(dentry)) {
1326 *ret = 1;
1327 return D_WALK_QUIT;
1328 }
1329 return D_WALK_CONTINUE;
1330}
1331
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001332/**
1333 * have_submounts - check for mounts over a dentry
1334 * @parent: dentry to check.
1335 *
1336 * Return true if the parent or its subdirectories contain
1337 * a mount point
1338 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001339int have_submounts(struct dentry *parent)
1340{
1341 int ret = 0;
1342
1343 d_walk(parent, &ret, check_mount, NULL);
1344
1345 return ret;
1346}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001347EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001350 * Called by mount code to set a mountpoint and check if the mountpoint is
1351 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1352 * subtree can become unreachable).
1353 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001354 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001355 * this reason take rename_lock and d_lock on dentry and ancestors.
1356 */
1357int d_set_mounted(struct dentry *dentry)
1358{
1359 struct dentry *p;
1360 int ret = -ENOENT;
1361 write_seqlock(&rename_lock);
1362 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001363 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001364 spin_lock(&p->d_lock);
1365 if (unlikely(d_unhashed(p))) {
1366 spin_unlock(&p->d_lock);
1367 goto out;
1368 }
1369 spin_unlock(&p->d_lock);
1370 }
1371 spin_lock(&dentry->d_lock);
1372 if (!d_unlinked(dentry)) {
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13001373 ret = -EBUSY;
1374 if (!d_mountpoint(dentry)) {
1375 dentry->d_flags |= DCACHE_MOUNTED;
1376 ret = 0;
1377 }
Miklos Szeredieed81002013-09-05 14:39:11 +02001378 }
1379 spin_unlock(&dentry->d_lock);
1380out:
1381 write_sequnlock(&rename_lock);
1382 return ret;
1383}
1384
1385/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001386 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 * and move any unused dentries to the end of the unused
1388 * list for prune_dcache(). We descend to the next level
1389 * whenever the d_subdirs list is non-empty and continue
1390 * searching.
1391 *
1392 * It returns zero iff there are no unused children,
1393 * otherwise it returns the number of children moved to
1394 * the end of the unused list. This may not be the total
1395 * number of unused children, because select_parent can
1396 * drop the lock and return early due to latency
1397 * constraints.
1398 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001399
1400struct select_data {
1401 struct dentry *start;
1402 struct list_head dispose;
1403 int found;
1404};
1405
1406static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001408 struct select_data *data = _data;
1409 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001411 if (data->start == dentry)
1412 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
Al Virofe915222014-05-03 00:02:25 -04001414 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001415 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001416 } else {
1417 if (dentry->d_flags & DCACHE_LRU_LIST)
1418 d_lru_del(dentry);
1419 if (!dentry->d_lockref.count) {
1420 d_shrink_add(dentry, &data->dispose);
1421 data->found++;
1422 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 }
1424 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001425 * We can return to the caller if we have found some (this
1426 * ensures forward progress). We'll be coming back to find
1427 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 */
Al Virofe915222014-05-03 00:02:25 -04001429 if (!list_empty(&data->dispose))
1430 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001432 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433}
1434
1435/**
1436 * shrink_dcache_parent - prune dcache
1437 * @parent: parent of entries to prune
1438 *
1439 * Prune the dcache to remove unused children of the parent dentry.
1440 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001441void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001443 for (;;) {
1444 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001446 INIT_LIST_HEAD(&data.dispose);
1447 data.start = parent;
1448 data.found = 0;
1449
1450 d_walk(parent, &data, select_collect, NULL);
1451 if (!data.found)
1452 break;
1453
1454 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001455 cond_resched();
1456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001458EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Al Viro9c8c10e2014-05-02 20:36:10 -04001460static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001461{
Al Viro9c8c10e2014-05-02 20:36:10 -04001462 /* it has busy descendents; complain about those instead */
1463 if (!list_empty(&dentry->d_subdirs))
1464 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001465
Al Viro9c8c10e2014-05-02 20:36:10 -04001466 /* root with refcount 1 is fine */
1467 if (dentry == _data && dentry->d_lockref.count == 1)
1468 return D_WALK_CONTINUE;
1469
1470 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1471 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001472 dentry,
1473 dentry->d_inode ?
1474 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001475 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001476 dentry->d_lockref.count,
1477 dentry->d_sb->s_type->name,
1478 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001479 WARN_ON(1);
1480 return D_WALK_CONTINUE;
1481}
1482
1483static void do_one_tree(struct dentry *dentry)
1484{
1485 shrink_dcache_parent(dentry);
1486 d_walk(dentry, dentry, umount_check, NULL);
1487 d_drop(dentry);
1488 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001489}
1490
1491/*
1492 * destroy the dentries attached to a superblock on unmounting
1493 */
1494void shrink_dcache_for_umount(struct super_block *sb)
1495{
1496 struct dentry *dentry;
1497
Al Viro9c8c10e2014-05-02 20:36:10 -04001498 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001499
1500 dentry = sb->s_root;
1501 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001502 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001503
1504 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001505 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1506 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001507 }
1508}
1509
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001510struct detach_data {
1511 struct select_data select;
1512 struct dentry *mountpoint;
1513};
1514static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001515{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001516 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001517
1518 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001519 __dget_dlock(dentry);
1520 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001521 return D_WALK_QUIT;
1522 }
1523
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001524 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001525}
1526
1527static void check_and_drop(void *_data)
1528{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001529 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001530
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001531 if (!data->mountpoint && !data->select.found)
1532 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001533}
1534
1535/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001536 * d_invalidate - detach submounts, prune dcache, and drop
1537 * @dentry: dentry to invalidate (aka detach, prune and drop)
1538 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001539 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001540 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001541 * The final d_drop is done as an atomic operation relative to
1542 * rename_lock ensuring there are no races with d_set_mounted. This
1543 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001544 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001545void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001546{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001547 /*
1548 * If it's already been dropped, return OK.
1549 */
1550 spin_lock(&dentry->d_lock);
1551 if (d_unhashed(dentry)) {
1552 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001553 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001554 }
1555 spin_unlock(&dentry->d_lock);
1556
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001557 /* Negative dentries can be dropped without further checks */
1558 if (!dentry->d_inode) {
1559 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001560 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001561 }
1562
1563 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001564 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001565
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001566 data.mountpoint = NULL;
1567 INIT_LIST_HEAD(&data.select.dispose);
1568 data.select.start = dentry;
1569 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001570
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001571 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001572
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001573 if (data.select.found)
1574 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001575
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001576 if (data.mountpoint) {
1577 detach_mounts(data.mountpoint);
1578 dput(data.mountpoint);
1579 }
1580
1581 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001582 break;
1583
1584 cond_resched();
1585 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001586}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001587EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589/**
Al Viroa4464db2011-07-07 15:03:58 -04001590 * __d_alloc - allocate a dcache entry
1591 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 * @name: qstr of the name
1593 *
1594 * Allocates a dentry. It returns %NULL if there is insufficient memory
1595 * available. On a success the dentry is returned. The name passed in is
1596 * copied and the copy passed in may be reused after this call.
1597 */
1598
Al Viroa4464db2011-07-07 15:03:58 -04001599struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600{
1601 struct dentry *dentry;
1602 char *dname;
Miklos Szeredi285b1022016-06-28 11:47:32 +02001603 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
Mel Gormane12ba742007-10-16 01:25:52 -07001605 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 if (!dentry)
1607 return NULL;
1608
Linus Torvalds6326c712012-05-21 16:14:04 -07001609 /*
1610 * We guarantee that the inline name is always NUL-terminated.
1611 * This way the memcpy() done by the name switching in rename
1612 * will still always have a NUL at the end, even if we might
1613 * be overwriting an internal NUL character
1614 */
1615 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001616 if (unlikely(!name)) {
1617 static const struct qstr anon = QSTR_INIT("/", 1);
1618 name = &anon;
1619 dname = dentry->d_iname;
1620 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001621 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001622 struct external_name *p = kmalloc(size + name->len,
1623 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001624 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 kmem_cache_free(dentry_cache, dentry);
1626 return NULL;
1627 }
Al Viro8d85b482014-09-29 14:54:27 -04001628 atomic_set(&p->u.count, 1);
1629 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001630 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1631 kasan_unpoison_shadow(dname,
1632 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 } else {
1634 dname = dentry->d_iname;
1635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 dentry->d_name.len = name->len;
1638 dentry->d_name.hash = name->hash;
1639 memcpy(dname, name->name, name->len);
1640 dname[name->len] = 0;
1641
Linus Torvalds6326c712012-05-21 16:14:04 -07001642 /* Make sure we always see the terminating NUL character */
1643 smp_wmb();
1644 dentry->d_name.name = dname;
1645
Waiman Long98474232013-08-28 18:24:59 -07001646 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001647 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001649 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001651 dentry->d_parent = dentry;
1652 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 dentry->d_op = NULL;
1654 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001655 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 INIT_LIST_HEAD(&dentry->d_lru);
1657 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001658 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1659 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001660 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Miklos Szeredi285b1022016-06-28 11:47:32 +02001662 if (dentry->d_op && dentry->d_op->d_init) {
1663 err = dentry->d_op->d_init(dentry);
1664 if (err) {
1665 if (dname_external(dentry))
1666 kfree(external_name(dentry));
1667 kmem_cache_free(dentry_cache, dentry);
1668 return NULL;
1669 }
1670 }
1671
Nick Piggin3e880fb2011-01-07 17:49:19 +11001672 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001673
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 return dentry;
1675}
Al Viroa4464db2011-07-07 15:03:58 -04001676
1677/**
1678 * d_alloc - allocate a dcache entry
1679 * @parent: parent of entry to allocate
1680 * @name: qstr of the name
1681 *
1682 * Allocates a dentry. It returns %NULL if there is insufficient memory
1683 * available. On a success the dentry is returned. The name passed in is
1684 * copied and the copy passed in may be reused after this call.
1685 */
1686struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1687{
1688 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1689 if (!dentry)
1690 return NULL;
Al Viro3d56c252016-06-07 21:26:55 -04001691 dentry->d_flags |= DCACHE_RCUACCESS;
Al Viroa4464db2011-07-07 15:03:58 -04001692 spin_lock(&parent->d_lock);
1693 /*
1694 * don't need child lock because it is not subject
1695 * to concurrency here
1696 */
1697 __dget_dlock(parent);
1698 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001699 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001700 spin_unlock(&parent->d_lock);
1701
1702 return dentry;
1703}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001704EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Al Viroba65dc52016-06-10 11:32:47 -04001706struct dentry *d_alloc_cursor(struct dentry * parent)
1707{
1708 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1709 if (dentry) {
1710 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1711 dentry->d_parent = dget(parent);
1712 }
1713 return dentry;
1714}
1715
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001716/**
1717 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1718 * @sb: the superblock
1719 * @name: qstr of the name
1720 *
1721 * For a filesystem that just pins its dentries in memory and never
1722 * performs lookups at all, return an unhashed IS_ROOT dentry.
1723 */
Nick Piggin4b936882011-01-07 17:50:07 +11001724struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1725{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001726 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001727}
1728EXPORT_SYMBOL(d_alloc_pseudo);
1729
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1731{
1732 struct qstr q;
1733
1734 q.name = name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07001735 q.hash_len = hashlen_string(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 return d_alloc(parent, &q);
1737}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001738EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
Nick Pigginfb045ad2011-01-07 17:49:55 +11001740void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1741{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001742 WARN_ON_ONCE(dentry->d_op);
1743 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001744 DCACHE_OP_COMPARE |
1745 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001746 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001747 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001748 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001749 dentry->d_op = op;
1750 if (!op)
1751 return;
1752 if (op->d_hash)
1753 dentry->d_flags |= DCACHE_OP_HASH;
1754 if (op->d_compare)
1755 dentry->d_flags |= DCACHE_OP_COMPARE;
1756 if (op->d_revalidate)
1757 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001758 if (op->d_weak_revalidate)
1759 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001760 if (op->d_delete)
1761 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001762 if (op->d_prune)
1763 dentry->d_flags |= DCACHE_OP_PRUNE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001764 if (op->d_real)
1765 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001766
1767}
1768EXPORT_SYMBOL(d_set_d_op);
1769
David Howellsdf1a0852015-01-29 12:02:28 +00001770
1771/*
1772 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1773 * @dentry - The dentry to mark
1774 *
1775 * Mark a dentry as falling through to the lower layer (as set with
1776 * d_pin_lower()). This flag may be recorded on the medium.
1777 */
1778void d_set_fallthru(struct dentry *dentry)
1779{
1780 spin_lock(&dentry->d_lock);
1781 dentry->d_flags |= DCACHE_FALLTHRU;
1782 spin_unlock(&dentry->d_lock);
1783}
1784EXPORT_SYMBOL(d_set_fallthru);
1785
David Howellsb18825a2013-09-12 19:22:53 +01001786static unsigned d_flags_for_inode(struct inode *inode)
1787{
David Howells44bdb5e2015-01-29 12:02:29 +00001788 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001789
1790 if (!inode)
1791 return DCACHE_MISS_TYPE;
1792
1793 if (S_ISDIR(inode->i_mode)) {
1794 add_flags = DCACHE_DIRECTORY_TYPE;
1795 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1796 if (unlikely(!inode->i_op->lookup))
1797 add_flags = DCACHE_AUTODIR_TYPE;
1798 else
1799 inode->i_opflags |= IOP_LOOKUP;
1800 }
David Howells44bdb5e2015-01-29 12:02:29 +00001801 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001802 }
1803
David Howells44bdb5e2015-01-29 12:02:29 +00001804 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001805 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001806 add_flags = DCACHE_SYMLINK_TYPE;
1807 goto type_determined;
1808 }
1809 inode->i_opflags |= IOP_NOFOLLOW;
1810 }
1811
1812 if (unlikely(!S_ISREG(inode->i_mode)))
1813 add_flags = DCACHE_SPECIAL_TYPE;
1814
1815type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001816 if (unlikely(IS_AUTOMOUNT(inode)))
1817 add_flags |= DCACHE_NEED_AUTOMOUNT;
1818 return add_flags;
1819}
1820
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001821static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1822{
David Howellsb18825a2013-09-12 19:22:53 +01001823 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001824 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001825
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001826 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001827 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001828 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001829 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001830 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04001831 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001832 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001833}
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835/**
1836 * d_instantiate - fill in inode information for a dentry
1837 * @entry: dentry to complete
1838 * @inode: inode to attach to this dentry
1839 *
1840 * Fill in inode information in the entry.
1841 *
1842 * This turns negative dentries into productive full members
1843 * of society.
1844 *
1845 * NOTE! This assumes that the inode count has been incremented
1846 * (or otherwise set) by the caller to indicate that it is now
1847 * in use by the dcache.
1848 */
1849
1850void d_instantiate(struct dentry *entry, struct inode * inode)
1851{
Al Viro946e51f2014-10-26 19:19:16 -04001852 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001853 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001854 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001855 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001856 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001857 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001858 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001860EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
Al Viro2d2d3f12018-05-04 08:23:01 -04001862/*
1863 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1864 * with lockdep-related part of unlock_new_inode() done before
1865 * anything else. Use that instead of open-coding d_instantiate()/
1866 * unlock_new_inode() combinations.
1867 */
1868void d_instantiate_new(struct dentry *entry, struct inode *inode)
1869{
1870 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1871 BUG_ON(!inode);
1872 lockdep_annotate_inode_mutex_key(inode);
1873 security_d_instantiate(entry, inode);
1874 spin_lock(&inode->i_lock);
1875 __d_instantiate(entry, inode);
1876 WARN_ON(!(inode->i_state & I_NEW));
1877 inode->i_state &= ~I_NEW;
1878 smp_mb();
1879 wake_up_bit(&inode->i_state, __I_NEW);
1880 spin_unlock(&inode->i_lock);
1881}
1882EXPORT_SYMBOL(d_instantiate_new);
1883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001885 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1886 * @entry: dentry to complete
1887 * @inode: inode to attach to this dentry
1888 *
1889 * Fill in inode information in the entry. If a directory alias is found, then
1890 * return an error (and drop inode). Together with d_materialise_unique() this
1891 * guarantees that a directory inode may never have more than one alias.
1892 */
1893int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1894{
Al Viro946e51f2014-10-26 19:19:16 -04001895 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001896
Al Virob9680912016-04-11 00:53:26 -04001897 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001898 spin_lock(&inode->i_lock);
1899 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1900 spin_unlock(&inode->i_lock);
1901 iput(inode);
1902 return -EBUSY;
1903 }
1904 __d_instantiate(entry, inode);
1905 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001906
1907 return 0;
1908}
1909EXPORT_SYMBOL(d_instantiate_no_diralias);
1910
Al Viroadc0e912012-01-08 16:49:21 -05001911struct dentry *d_make_root(struct inode *root_inode)
1912{
1913 struct dentry *res = NULL;
1914
1915 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001916 res = __d_alloc(root_inode->i_sb, NULL);
Al Viroadc0e912012-01-08 16:49:21 -05001917 if (res)
1918 d_instantiate(res, root_inode);
1919 else
1920 iput(root_inode);
1921 }
1922 return res;
1923}
1924EXPORT_SYMBOL(d_make_root);
1925
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001926static struct dentry * __d_find_any_alias(struct inode *inode)
1927{
1928 struct dentry *alias;
1929
Al Virob3d9b7a2012-06-09 13:51:19 -04001930 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001931 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001932 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001933 __dget(alias);
1934 return alias;
1935}
1936
Sage Weil46f72b32012-01-10 09:04:37 -08001937/**
1938 * d_find_any_alias - find any alias for a given inode
1939 * @inode: inode to find an alias for
1940 *
1941 * If any aliases exist for the given inode, take and return a
1942 * reference for one of them. If no aliases exist, return %NULL.
1943 */
1944struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001945{
1946 struct dentry *de;
1947
1948 spin_lock(&inode->i_lock);
1949 de = __d_find_any_alias(inode);
1950 spin_unlock(&inode->i_lock);
1951 return de;
1952}
Sage Weil46f72b32012-01-10 09:04:37 -08001953EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001954
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001955static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001956{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001957 struct dentry *tmp;
1958 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001959 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001960
1961 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001962 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001963 if (IS_ERR(inode))
1964 return ERR_CAST(inode);
1965
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001966 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001967 if (res)
1968 goto out_iput;
1969
Al Viro798434b2016-03-24 20:38:43 -04001970 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001971 if (!tmp) {
1972 res = ERR_PTR(-ENOMEM);
1973 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001974 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001975
Al Virob9680912016-04-11 00:53:26 -04001976 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001977 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001978 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001979 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001980 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001981 dput(tmp);
1982 goto out_iput;
1983 }
1984
1985 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001986 add_flags = d_flags_for_inode(inode);
1987
1988 if (disconnected)
1989 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001990
Christoph Hellwig9308a612008-08-11 15:49:12 +02001991 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001992 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001993 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001994 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001995 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001996 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001997 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001998 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001999
Christoph Hellwig9308a612008-08-11 15:49:12 +02002000 return tmp;
2001
2002 out_iput:
2003 iput(inode);
2004 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002005}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002006
2007/**
2008 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2009 * @inode: inode to allocate the dentry for
2010 *
2011 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2012 * similar open by handle operations. The returned dentry may be anonymous,
2013 * or may have a full name (if the inode was already in the cache).
2014 *
2015 * When called on a directory inode, we must ensure that the inode only ever
2016 * has one dentry. If a dentry is found, that is returned instead of
2017 * allocating a new one.
2018 *
2019 * On successful return, the reference to the inode has been transferred
2020 * to the dentry. In case of an error the reference on the inode is released.
2021 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2022 * be passed in and the error will be propagated to the return value,
2023 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2024 */
2025struct dentry *d_obtain_alias(struct inode *inode)
2026{
2027 return __d_obtain_alias(inode, 1);
2028}
Benny Halevyadc48722009-02-27 14:02:59 -08002029EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030
2031/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002032 * d_obtain_root - find or allocate a dentry for a given inode
2033 * @inode: inode to allocate the dentry for
2034 *
2035 * Obtain an IS_ROOT dentry for the root of a filesystem.
2036 *
2037 * We must ensure that directory inodes only ever have one dentry. If a
2038 * dentry is found, that is returned instead of allocating a new one.
2039 *
2040 * On successful return, the reference to the inode has been transferred
2041 * to the dentry. In case of an error the reference on the inode is
2042 * released. A %NULL or IS_ERR inode may be passed in and will be the
2043 * error will be propagate to the return value, with a %NULL @inode
2044 * replaced by ERR_PTR(-ESTALE).
2045 */
2046struct dentry *d_obtain_root(struct inode *inode)
2047{
2048 return __d_obtain_alias(inode, 0);
2049}
2050EXPORT_SYMBOL(d_obtain_root);
2051
2052/**
Barry Naujok94035402008-05-21 16:50:46 +10002053 * d_add_ci - lookup or allocate new dentry with case-exact name
2054 * @inode: the inode case-insensitive lookup has found
2055 * @dentry: the negative dentry that was passed to the parent's lookup func
2056 * @name: the case-exact name to be associated with the returned dentry
2057 *
2058 * This is to avoid filling the dcache with case-insensitive names to the
2059 * same inode, only the actual correct case is stored in the dcache for
2060 * case-insensitive filesystems.
2061 *
2062 * For a case-insensitive lookup match and if the the case-exact dentry
2063 * already exists in in the dcache, use it and return it.
2064 *
2065 * If no entry exists with the exact case name, allocate new dentry with
2066 * the exact case, and return the spliced entry.
2067 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002068struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002069 struct qstr *name)
2070{
Al Virod9171b92016-04-15 03:33:13 -04002071 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002072
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002073 /*
2074 * First check if a dentry matching the name already exists,
2075 * if not go ahead and create it now.
2076 */
Barry Naujok94035402008-05-21 16:50:46 +10002077 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002078 if (found) {
2079 iput(inode);
2080 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002081 }
Al Virod9171b92016-04-15 03:33:13 -04002082 if (d_in_lookup(dentry)) {
2083 found = d_alloc_parallel(dentry->d_parent, name,
2084 dentry->d_wait);
2085 if (IS_ERR(found) || !d_in_lookup(found)) {
2086 iput(inode);
2087 return found;
2088 }
2089 } else {
2090 found = d_alloc(dentry->d_parent, name);
2091 if (!found) {
2092 iput(inode);
2093 return ERR_PTR(-ENOMEM);
2094 }
2095 }
2096 res = d_splice_alias(inode, found);
2097 if (res) {
2098 dput(found);
2099 return res;
2100 }
Al Viro4f522a22013-02-11 23:20:37 -05002101 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002102}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002103EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002105
Al Virod4c91a82016-06-25 23:33:49 -04002106static inline bool d_same_name(const struct dentry *dentry,
2107 const struct dentry *parent,
2108 const struct qstr *name)
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002109{
Al Virod4c91a82016-06-25 23:33:49 -04002110 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2111 if (dentry->d_name.len != name->len)
2112 return false;
2113 return dentry_cmp(dentry, name->name, name->len) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002114 }
Al Viro6fa67e72016-07-31 16:37:25 -04002115 return parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002116 dentry->d_name.len, dentry->d_name.name,
2117 name) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002118}
2119
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002121 * __d_lookup_rcu - search for a dentry (racy, store-free)
2122 * @parent: parent dentry
2123 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002124 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002125 * Returns: dentry, or NULL
2126 *
2127 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2128 * resolution (store-free path walking) design described in
2129 * Documentation/filesystems/path-lookup.txt.
2130 *
2131 * This is not to be used outside core vfs.
2132 *
2133 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2134 * held, and rcu_read_lock held. The returned dentry must not be stored into
2135 * without taking d_lock and checking d_seq sequence count against @seq
2136 * returned here.
2137 *
Linus Torvalds15570082013-09-02 11:38:06 -07002138 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002139 * function.
2140 *
2141 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2142 * the returned dentry, so long as its parent's seqlock is checked after the
2143 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2144 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002145 *
2146 * NOTE! The caller *has* to check the resulting dentry against the sequence
2147 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002148 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002149struct dentry *__d_lookup_rcu(const struct dentry *parent,
2150 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002151 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002152{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002153 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002154 const unsigned char *str = name->name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002155 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002156 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002157 struct dentry *dentry;
2158
2159 /*
2160 * Note: There is significant duplication with __d_lookup_rcu which is
2161 * required to prevent single threaded performance regressions
2162 * especially on architectures where smp_rmb (in seqcounts) are costly.
2163 * Keep the two functions in sync.
2164 */
2165
2166 /*
2167 * The hash list is protected using RCU.
2168 *
2169 * Carefully use d_seq when comparing a candidate dentry, to avoid
2170 * races with d_move().
2171 *
2172 * It is possible that concurrent renames can mess up our list
2173 * walk here and result in missing our dentry, resulting in the
2174 * false-negative result. d_lookup() protects against concurrent
2175 * renames using rename_lock seqlock.
2176 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002177 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002178 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002179 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002180 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002181
Nick Piggin31e6b012011-01-07 17:49:52 +11002182seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002183 /*
2184 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002185 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002186 *
2187 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002188 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002189 *
2190 * NOTE! We do a "raw" seqcount_begin here. That means that
2191 * we don't wait for the sequence count to stabilize if it
2192 * is in the middle of a sequence change. If we do the slow
2193 * dentry compare, we will do seqretries until it is stable,
2194 * and if we end up with a successful lookup, we actually
2195 * want to exit RCU lookup anyway.
Al Virod4c91a82016-06-25 23:33:49 -04002196 *
2197 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2198 * we are still guaranteed NUL-termination of ->d_name.name.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002199 */
2200 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002201 if (dentry->d_parent != parent)
2202 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002203 if (d_unhashed(dentry))
2204 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002205
2206 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Al Virod4c91a82016-06-25 23:33:49 -04002207 int tlen;
2208 const char *tname;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002209 if (dentry->d_name.hash != hashlen_hash(hashlen))
2210 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002211 tlen = dentry->d_name.len;
2212 tname = dentry->d_name.name;
2213 /* we want a consistent (name,len) pair */
2214 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2215 cpu_relax();
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002216 goto seqretry;
2217 }
Al Viro6fa67e72016-07-31 16:37:25 -04002218 if (parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002219 tlen, tname, name) != 0)
2220 continue;
2221 } else {
2222 if (dentry->d_name.hash_len != hashlen)
2223 continue;
2224 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2225 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002226 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002227 *seqp = seq;
Al Virod4c91a82016-06-25 23:33:49 -04002228 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002229 }
2230 return NULL;
2231}
2232
2233/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 * d_lookup - search for a dentry
2235 * @parent: parent dentry
2236 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002237 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002239 * d_lookup searches the children of the parent dentry for the name in
2240 * question. If the dentry is found its reference count is incremented and the
2241 * dentry is returned. The caller must use dput to free the entry when it has
2242 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 */
Al Viroda2d8452013-01-24 18:29:34 -05002244struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245{
Nick Piggin31e6b012011-01-07 17:49:52 +11002246 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002247 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
Daeseok Younb8314f92014-08-11 11:46:53 +09002249 do {
2250 seq = read_seqbegin(&rename_lock);
2251 dentry = __d_lookup(parent, name);
2252 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 break;
2254 } while (read_seqretry(&rename_lock, seq));
2255 return dentry;
2256}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002257EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258
Nick Piggin31e6b012011-01-07 17:49:52 +11002259/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002260 * __d_lookup - search for a dentry (racy)
2261 * @parent: parent dentry
2262 * @name: qstr of name we wish to find
2263 * Returns: dentry, or NULL
2264 *
2265 * __d_lookup is like d_lookup, however it may (rarely) return a
2266 * false-negative result due to unrelated rename activity.
2267 *
2268 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2269 * however it must be used carefully, eg. with a following d_lookup in
2270 * the case of failure.
2271 *
2272 * __d_lookup callers must be commented.
2273 */
Al Viroa713ca22013-01-24 18:27:00 -05002274struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 unsigned int hash = name->hash;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002277 struct hlist_bl_head *b = d_hash(hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002278 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002279 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002280 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
Nick Pigginb04f7842010-08-18 04:37:34 +10002282 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002283 * Note: There is significant duplication with __d_lookup_rcu which is
2284 * required to prevent single threaded performance regressions
2285 * especially on architectures where smp_rmb (in seqcounts) are costly.
2286 * Keep the two functions in sync.
2287 */
2288
2289 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002290 * The hash list is protected using RCU.
2291 *
2292 * Take d_lock when comparing a candidate dentry, to avoid races
2293 * with d_move().
2294 *
2295 * It is possible that concurrent renames can mess up our list
2296 * walk here and result in missing our dentry, resulting in the
2297 * false-negative result. d_lookup() protects against concurrent
2298 * renames using rename_lock seqlock.
2299 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002300 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002301 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 rcu_read_lock();
2303
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002304 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 if (dentry->d_name.hash != hash)
2307 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 if (dentry->d_parent != parent)
2311 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002312 if (d_unhashed(dentry))
2313 goto next;
2314
Al Virod4c91a82016-06-25 23:33:49 -04002315 if (!d_same_name(dentry, parent, name))
2316 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
Waiman Long98474232013-08-28 18:24:59 -07002318 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002319 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 spin_unlock(&dentry->d_lock);
2321 break;
2322next:
2323 spin_unlock(&dentry->d_lock);
2324 }
2325 rcu_read_unlock();
2326
2327 return found;
2328}
2329
2330/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002331 * d_hash_and_lookup - hash the qstr then search for a dentry
2332 * @dir: Directory to search in
2333 * @name: qstr of name we wish to find
2334 *
Al Viro4f522a22013-02-11 23:20:37 -05002335 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002336 */
2337struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2338{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002339 /*
2340 * Check for a fs-specific hash function. Note that we must
2341 * calculate the standard hash first, as the d_op->d_hash()
2342 * routine may choose to leave the hash value unchanged.
2343 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002344 name->hash = full_name_hash(dir, name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002345 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002346 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002347 if (unlikely(err < 0))
2348 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002349 }
Al Viro4f522a22013-02-11 23:20:37 -05002350 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002351}
Al Viro4f522a22013-02-11 23:20:37 -05002352EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002353
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354/*
2355 * When a file is deleted, we have two options:
2356 * - turn this dentry into a negative dentry
2357 * - unhash this dentry and free it.
2358 *
2359 * Usually, we want to just turn this into
2360 * a negative dentry, but if anybody else is
2361 * currently using the dentry or the inode
2362 * we can't do that and we fall back on removing
2363 * it from the hash queues and waiting for
2364 * it to be deleted later when it has no users
2365 */
2366
2367/**
2368 * d_delete - delete a dentry
2369 * @dentry: The dentry to delete
2370 *
2371 * Turn the dentry into a negative dentry if possible, otherwise
2372 * remove it from the hash queues so it can be deleted later
2373 */
2374
2375void d_delete(struct dentry * dentry)
2376{
Nick Piggin873feea2011-01-07 17:50:06 +11002377 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002378 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 /*
2380 * Are we the only user?
2381 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002382again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002384 inode = dentry->d_inode;
2385 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002386 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002387 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002388 spin_unlock(&dentry->d_lock);
2389 cpu_relax();
2390 goto again;
2391 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002392 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002393 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002394 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return;
2396 }
2397
2398 if (!d_unhashed(dentry))
2399 __d_drop(dentry);
2400
2401 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002402
2403 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002405EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
Al Viro15d3c582016-07-29 17:45:21 -04002407static void __d_rehash(struct dentry *entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
Al Viro15d3c582016-07-29 17:45:21 -04002409 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
NeilBrown3aa66ba2017-11-10 15:45:41 +11002410
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002411 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002412 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002413 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414}
2415
2416/**
2417 * d_rehash - add an entry back to the hash
2418 * @entry: dentry to add to the hash
2419 *
2420 * Adds a dentry to the hash according to its name.
2421 */
2422
2423void d_rehash(struct dentry * entry)
2424{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 spin_lock(&entry->d_lock);
Al Viro15d3c582016-07-29 17:45:21 -04002426 __d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002429EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Al Viro84e710d2016-04-15 00:58:55 -04002431static inline unsigned start_dir_add(struct inode *dir)
2432{
2433
2434 for (;;) {
2435 unsigned n = dir->i_dir_seq;
2436 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2437 return n;
2438 cpu_relax();
2439 }
2440}
2441
2442static inline void end_dir_add(struct inode *dir, unsigned n)
2443{
2444 smp_store_release(&dir->i_dir_seq, n + 2);
2445}
2446
Al Virod9171b92016-04-15 03:33:13 -04002447static void d_wait_lookup(struct dentry *dentry)
2448{
2449 if (d_in_lookup(dentry)) {
2450 DECLARE_WAITQUEUE(wait, current);
2451 add_wait_queue(dentry->d_wait, &wait);
2452 do {
2453 set_current_state(TASK_UNINTERRUPTIBLE);
2454 spin_unlock(&dentry->d_lock);
2455 schedule();
2456 spin_lock(&dentry->d_lock);
2457 } while (d_in_lookup(dentry));
2458 }
2459}
2460
Al Viro94bdd652016-04-15 02:42:04 -04002461struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002462 const struct qstr *name,
2463 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002464{
Al Viro94bdd652016-04-15 02:42:04 -04002465 unsigned int hash = name->hash;
Al Viro94bdd652016-04-15 02:42:04 -04002466 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2467 struct hlist_bl_node *node;
2468 struct dentry *new = d_alloc(parent, name);
2469 struct dentry *dentry;
2470 unsigned seq, r_seq, d_seq;
2471
2472 if (unlikely(!new))
2473 return ERR_PTR(-ENOMEM);
2474
2475retry:
2476 rcu_read_lock();
Will Deaconbcefedb2018-02-19 14:55:54 +00002477 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
Al Viro94bdd652016-04-15 02:42:04 -04002478 r_seq = read_seqbegin(&rename_lock);
2479 dentry = __d_lookup_rcu(parent, name, &d_seq);
2480 if (unlikely(dentry)) {
2481 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2482 rcu_read_unlock();
2483 goto retry;
2484 }
2485 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2486 rcu_read_unlock();
2487 dput(dentry);
2488 goto retry;
2489 }
2490 rcu_read_unlock();
2491 dput(new);
2492 return dentry;
2493 }
2494 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2495 rcu_read_unlock();
2496 goto retry;
2497 }
Will Deaconbcefedb2018-02-19 14:55:54 +00002498
2499 if (unlikely(seq & 1)) {
2500 rcu_read_unlock();
2501 goto retry;
2502 }
2503
Al Viro94bdd652016-04-15 02:42:04 -04002504 hlist_bl_lock(b);
Will Deacon527ed412018-02-19 14:55:55 +00002505 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
Al Viro94bdd652016-04-15 02:42:04 -04002506 hlist_bl_unlock(b);
2507 rcu_read_unlock();
2508 goto retry;
2509 }
Al Viro94bdd652016-04-15 02:42:04 -04002510 /*
2511 * No changes for the parent since the beginning of d_lookup().
2512 * Since all removals from the chain happen with hlist_bl_lock(),
2513 * any potential in-lookup matches are going to stay here until
2514 * we unlock the chain. All fields are stable in everything
2515 * we encounter.
2516 */
2517 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2518 if (dentry->d_name.hash != hash)
2519 continue;
2520 if (dentry->d_parent != parent)
2521 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002522 if (!d_same_name(dentry, parent, name))
2523 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002524 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002525 /* now we can try to grab a reference */
2526 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2527 rcu_read_unlock();
2528 goto retry;
2529 }
2530
2531 rcu_read_unlock();
2532 /*
2533 * somebody is likely to be still doing lookup for it;
2534 * wait for them to finish
2535 */
Al Virod9171b92016-04-15 03:33:13 -04002536 spin_lock(&dentry->d_lock);
2537 d_wait_lookup(dentry);
2538 /*
2539 * it's not in-lookup anymore; in principle we should repeat
2540 * everything from dcache lookup, but it's likely to be what
2541 * d_lookup() would've found anyway. If it is, just return it;
2542 * otherwise we really have to repeat the whole thing.
2543 */
2544 if (unlikely(dentry->d_name.hash != hash))
2545 goto mismatch;
2546 if (unlikely(dentry->d_parent != parent))
2547 goto mismatch;
2548 if (unlikely(d_unhashed(dentry)))
2549 goto mismatch;
Al Virod4c91a82016-06-25 23:33:49 -04002550 if (unlikely(!d_same_name(dentry, parent, name)))
2551 goto mismatch;
Al Virod9171b92016-04-15 03:33:13 -04002552 /* OK, it *is* a hashed match; return it */
2553 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002554 dput(new);
2555 return dentry;
2556 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002557 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002558 /* we can't take ->d_lock here; it's OK, though. */
2559 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002560 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002561 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2562 hlist_bl_unlock(b);
2563 return new;
Al Virod9171b92016-04-15 03:33:13 -04002564mismatch:
2565 spin_unlock(&dentry->d_lock);
2566 dput(dentry);
2567 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002568}
2569EXPORT_SYMBOL(d_alloc_parallel);
2570
Al Viro85c7f812016-04-14 19:52:13 -04002571void __d_lookup_done(struct dentry *dentry)
2572{
Al Viro94bdd652016-04-15 02:42:04 -04002573 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2574 dentry->d_name.hash);
2575 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002576 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002577 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002578 wake_up_all(dentry->d_wait);
2579 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002580 hlist_bl_unlock(b);
2581 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002582 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002583}
2584EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002585
2586/* inode->i_lock held if inode is non-NULL */
2587
2588static inline void __d_add(struct dentry *dentry, struct inode *inode)
2589{
Al Viro84e710d2016-04-15 00:58:55 -04002590 struct inode *dir = NULL;
2591 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002592 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002593 if (unlikely(d_in_lookup(dentry))) {
2594 dir = dentry->d_parent->d_inode;
2595 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002596 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002597 }
Al Viroed782b52016-03-09 19:52:39 -05002598 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002599 unsigned add_flags = d_flags_for_inode(inode);
2600 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2601 raw_write_seqcount_begin(&dentry->d_seq);
2602 __d_set_inode_and_type(dentry, inode, add_flags);
2603 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002604 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002605 }
Al Viro15d3c582016-07-29 17:45:21 -04002606 __d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002607 if (dir)
2608 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002609 spin_unlock(&dentry->d_lock);
2610 if (inode)
2611 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002612}
2613
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002614/**
Al Viro34d0d192016-03-08 21:01:03 -05002615 * d_add - add dentry to hash queues
2616 * @entry: dentry to add
2617 * @inode: The inode to attach to this dentry
2618 *
2619 * This adds the entry to the hash queues and initializes @inode.
2620 * The entry was actually filled in earlier during d_alloc().
2621 */
2622
2623void d_add(struct dentry *entry, struct inode *inode)
2624{
Al Virob9680912016-04-11 00:53:26 -04002625 if (inode) {
2626 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002627 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002628 }
Al Viroed782b52016-03-09 19:52:39 -05002629 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002630}
2631EXPORT_SYMBOL(d_add);
2632
2633/**
Al Viro668d0cd2016-03-08 12:44:17 -05002634 * d_exact_alias - find and hash an exact unhashed alias
2635 * @entry: dentry to add
2636 * @inode: The inode to go with this dentry
2637 *
2638 * If an unhashed dentry with the same name/parent and desired
2639 * inode already exists, hash and return it. Otherwise, return
2640 * NULL.
2641 *
2642 * Parent directory should be locked.
2643 */
2644struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2645{
2646 struct dentry *alias;
Al Viro668d0cd2016-03-08 12:44:17 -05002647 unsigned int hash = entry->d_name.hash;
2648
2649 spin_lock(&inode->i_lock);
2650 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2651 /*
2652 * Don't need alias->d_lock here, because aliases with
2653 * d_parent == entry->d_parent are not subject to name or
2654 * parent changes, because the parent inode i_mutex is held.
2655 */
2656 if (alias->d_name.hash != hash)
2657 continue;
2658 if (alias->d_parent != entry->d_parent)
2659 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002660 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
Al Viro668d0cd2016-03-08 12:44:17 -05002661 continue;
2662 spin_lock(&alias->d_lock);
2663 if (!d_unhashed(alias)) {
2664 spin_unlock(&alias->d_lock);
2665 alias = NULL;
2666 } else {
2667 __dget_dlock(alias);
Al Viro15d3c582016-07-29 17:45:21 -04002668 __d_rehash(alias);
Al Viro668d0cd2016-03-08 12:44:17 -05002669 spin_unlock(&alias->d_lock);
2670 }
2671 spin_unlock(&inode->i_lock);
2672 return alias;
2673 }
2674 spin_unlock(&inode->i_lock);
2675 return NULL;
2676}
2677EXPORT_SYMBOL(d_exact_alias);
2678
2679/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002680 * dentry_update_name_case - update case insensitive dentry with a new name
2681 * @dentry: dentry to be updated
2682 * @name: new name
2683 *
2684 * Update a case insensitive dentry with new case of name.
2685 *
2686 * dentry must have been returned by d_lookup with name @name. Old and new
2687 * name lengths must match (ie. no d_compare which allows mismatched name
2688 * lengths).
2689 *
2690 * Parent inode i_mutex must be held over d_lookup and into this call (to
2691 * keep renames and concurrent inserts, and readdir(2) away).
2692 */
Al Viro9aba36d2016-07-20 22:28:45 -04002693void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002694{
Al Viro59551022016-01-22 15:40:57 -05002695 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002696 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2697
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002698 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002699 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002700 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002701 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002702 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002703}
2704EXPORT_SYMBOL(dentry_update_name_case);
2705
Al Viro8d85b482014-09-29 14:54:27 -04002706static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707{
Al Viro8d85b482014-09-29 14:54:27 -04002708 if (unlikely(dname_external(target))) {
2709 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 /*
2711 * Both external: swap the pointers
2712 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002713 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 } else {
2715 /*
2716 * dentry:internal, target:external. Steal target's
2717 * storage and make target internal.
2718 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002719 memcpy(target->d_iname, dentry->d_name.name,
2720 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 dentry->d_name.name = target->d_name.name;
2722 target->d_name.name = target->d_iname;
2723 }
2724 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002725 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 /*
2727 * dentry:external, target:internal. Give dentry's
2728 * storage to target and make dentry internal
2729 */
2730 memcpy(dentry->d_iname, target->d_name.name,
2731 target->d_name.len + 1);
2732 target->d_name.name = dentry->d_name.name;
2733 dentry->d_name.name = dentry->d_iname;
2734 } else {
2735 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002736 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002738 unsigned int i;
2739 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002740 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2741 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002742 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2743 swap(((long *) &dentry->d_iname)[i],
2744 ((long *) &target->d_iname)[i]);
2745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 }
2747 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002748 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749}
2750
Al Viro8d85b482014-09-29 14:54:27 -04002751static void copy_name(struct dentry *dentry, struct dentry *target)
2752{
2753 struct external_name *old_name = NULL;
2754 if (unlikely(dname_external(dentry)))
2755 old_name = external_name(dentry);
2756 if (unlikely(dname_external(target))) {
2757 atomic_inc(&external_name(target)->u.count);
2758 dentry->d_name = target->d_name;
2759 } else {
2760 memcpy(dentry->d_iname, target->d_name.name,
2761 target->d_name.len + 1);
2762 dentry->d_name.name = dentry->d_iname;
2763 dentry->d_name.hash_len = target->d_name.hash_len;
2764 }
2765 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2766 kfree_rcu(old_name, u.head);
2767}
2768
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002769static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2770{
2771 /*
2772 * XXXX: do we really need to take target->d_lock?
2773 */
2774 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2775 spin_lock(&target->d_parent->d_lock);
2776 else {
2777 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2778 spin_lock(&dentry->d_parent->d_lock);
2779 spin_lock_nested(&target->d_parent->d_lock,
2780 DENTRY_D_LOCK_NESTED);
2781 } else {
2782 spin_lock(&target->d_parent->d_lock);
2783 spin_lock_nested(&dentry->d_parent->d_lock,
2784 DENTRY_D_LOCK_NESTED);
2785 }
2786 }
2787 if (target < dentry) {
2788 spin_lock_nested(&target->d_lock, 2);
2789 spin_lock_nested(&dentry->d_lock, 3);
2790 } else {
2791 spin_lock_nested(&dentry->d_lock, 2);
2792 spin_lock_nested(&target->d_lock, 3);
2793 }
2794}
2795
Al Viro986c0192014-09-26 23:11:15 -04002796static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002797{
2798 if (target->d_parent != dentry->d_parent)
2799 spin_unlock(&dentry->d_parent->d_lock);
2800 if (target->d_parent != target)
2801 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002802 spin_unlock(&target->d_lock);
2803 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002804}
2805
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002807 * When switching names, the actual string doesn't strictly have to
2808 * be preserved in the target - because we're dropping the target
2809 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002810 * the new name before we switch, unless we are going to rehash
2811 * it. Note that if we *do* unhash the target, we are not allowed
2812 * to rehash it without giving it a new name/hash key - whether
2813 * we swap or overwrite the names here, resulting name won't match
2814 * the reality in filesystem; it's only there for d_path() purposes.
2815 * Note that all of this is happening under rename_lock, so the
2816 * any hash lookup seeing it in the middle of manipulations will
2817 * be discarded anyway. So we do not care what happens to the hash
2818 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002820/*
Al Viro18367502011-07-12 21:42:24 -04002821 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 * @dentry: entry to move
2823 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002824 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 *
2826 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002827 * dcache entries should not be moved in this way. Caller must hold
2828 * rename_lock, the i_mutex of the source and target directories,
2829 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002831static void __d_move(struct dentry *dentry, struct dentry *target,
2832 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833{
Al Viro84e710d2016-04-15 00:58:55 -04002834 struct inode *dir = NULL;
2835 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 if (!dentry->d_inode)
2837 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2838
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002839 BUG_ON(d_ancestor(dentry, target));
2840 BUG_ON(d_ancestor(target, dentry));
2841
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002842 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002843 if (unlikely(d_in_lookup(target))) {
2844 dir = target->d_parent->d_inode;
2845 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002846 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848
Nick Piggin31e6b012011-01-07 17:49:52 +11002849 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002850 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002851
Al Viro15d3c582016-07-29 17:45:21 -04002852 /* unhash both */
NeilBrown3aa66ba2017-11-10 15:45:41 +11002853 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2854 ___d_drop(dentry);
2855 ___d_drop(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002858 if (exchange)
2859 swap_names(dentry, target);
2860 else
2861 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862
Al Viro15d3c582016-07-29 17:45:21 -04002863 /* rehash in new place(s) */
2864 __d_rehash(dentry);
2865 if (exchange)
2866 __d_rehash(target);
NeilBrown3aa66ba2017-11-10 15:45:41 +11002867 else
2868 target->d_hash.pprev = NULL;
Al Viro15d3c582016-07-29 17:45:21 -04002869
Al Viro63cf4272014-09-26 23:06:14 -04002870 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002872 /* splicing a tree */
Al Viro3d56c252016-06-07 21:26:55 -04002873 dentry->d_flags |= DCACHE_RCUACCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 dentry->d_parent = target->d_parent;
2875 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002876 list_del_init(&target->d_child);
2877 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002879 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002880 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002881 list_move(&target->d_child, &target->d_parent->d_subdirs);
2882 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002883 if (exchange)
Al Viroaffda482016-05-29 18:35:12 -04002884 fsnotify_update_flags(target);
2885 fsnotify_update_flags(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886 }
2887
Nick Piggin31e6b012011-01-07 17:49:52 +11002888 write_seqcount_end(&target->d_seq);
2889 write_seqcount_end(&dentry->d_seq);
2890
Al Viro84e710d2016-04-15 00:58:55 -04002891 if (dir)
2892 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002893 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002894}
2895
2896/*
2897 * d_move - move a dentry
2898 * @dentry: entry to move
2899 * @target: new dentry
2900 *
2901 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002902 * dcache entries should not be moved in this way. See the locking
2903 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002904 */
2905void d_move(struct dentry *dentry, struct dentry *target)
2906{
2907 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002908 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002910}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002911EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912
Miklos Szeredida1ce062014-04-01 17:08:43 +02002913/*
2914 * d_exchange - exchange two dentries
2915 * @dentry1: first dentry
2916 * @dentry2: second dentry
2917 */
2918void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2919{
2920 write_seqlock(&rename_lock);
2921
2922 WARN_ON(!dentry1->d_inode);
2923 WARN_ON(!dentry2->d_inode);
2924 WARN_ON(IS_ROOT(dentry1));
2925 WARN_ON(IS_ROOT(dentry2));
2926
2927 __d_move(dentry1, dentry2, true);
2928
2929 write_sequnlock(&rename_lock);
2930}
2931
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002932/**
2933 * d_ancestor - search for an ancestor
2934 * @p1: ancestor dentry
2935 * @p2: child dentry
2936 *
2937 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2938 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002939 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002940struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002941{
2942 struct dentry *p;
2943
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002944 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002945 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002946 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002947 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002948 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002949}
2950
2951/*
2952 * This helper attempts to cope with remotely renamed directories
2953 *
2954 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002955 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002956 *
2957 * Note: If ever the locking in lock_rename() changes, then please
2958 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002959 */
Al Virob5ae6b12014-10-12 22:16:02 -04002960static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002961 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002962{
Al Viro9902af72016-04-15 15:08:36 -04002963 struct mutex *m1 = NULL;
2964 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002965 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002966
2967 /* If alias and dentry share a parent, then no extra locks required */
2968 if (alias->d_parent == dentry->d_parent)
2969 goto out_unalias;
2970
Trond Myklebust9eaef272006-10-21 10:24:20 -07002971 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002972 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2973 goto out_err;
2974 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04002975 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002976 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04002977 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002978out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002979 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002980 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002981out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002982 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04002983 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002984 if (m1)
2985 mutex_unlock(m1);
2986 return ret;
2987}
2988
David Howells770bfad2006-08-22 20:06:07 -04002989/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002990 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2991 * @inode: the inode which may have a disconnected dentry
2992 * @dentry: a negative dentry which we want to point to the inode.
2993 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002994 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2995 * place of the given dentry and return it, else simply d_add the inode
2996 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002997 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002998 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2999 * we should error out: directories can't have multiple aliases.
3000 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003001 * This is needed in the lookup routine of any filesystem that is exportable
3002 * (via knfsd) so that we can build dcache paths to directories effectively.
3003 *
3004 * If a dentry was found and moved, then it is returned. Otherwise NULL
3005 * is returned. This matches the expected return value of ->lookup.
3006 *
3007 * Cluster filesystems may call this function with a negative, hashed dentry.
3008 * In that case, we know that the inode will be a regular file, and also this
3009 * will only occur during atomic_open. So we need to check for the dentry
3010 * being already hashed only in the final case.
3011 */
3012struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3013{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003014 if (IS_ERR(inode))
3015 return ERR_CAST(inode);
3016
David Howells770bfad2006-08-22 20:06:07 -04003017 BUG_ON(!d_unhashed(dentry));
3018
Al Virode689f52016-03-09 18:05:42 -05003019 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04003020 goto out;
Al Virode689f52016-03-09 18:05:42 -05003021
Al Virob9680912016-04-11 00:53:26 -04003022 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11003023 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003024 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04003025 struct dentry *new = __d_find_any_alias(inode);
3026 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05003027 /* The reference to new ensures it remains an alias */
3028 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04003029 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003030 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04003031 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003032 dput(new);
3033 new = ERR_PTR(-ELOOP);
3034 pr_warn_ratelimited(
3035 "VFS: Lookup of '%s' in %s %s"
3036 " would have caused loop\n",
3037 dentry->d_name.name,
3038 inode->i_sb->s_type->name,
3039 inode->i_sb->s_id);
3040 } else if (!IS_ROOT(new)) {
3041 int err = __d_unalias(inode, dentry, new);
3042 write_sequnlock(&rename_lock);
3043 if (err) {
3044 dput(new);
3045 new = ERR_PTR(err);
3046 }
Al Viro18367502011-07-12 21:42:24 -04003047 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003048 __d_move(new, dentry, false);
3049 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003050 }
Al Virob5ae6b12014-10-12 22:16:02 -04003051 iput(inode);
3052 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003053 }
David Howells770bfad2006-08-22 20:06:07 -04003054 }
Al Virob5ae6b12014-10-12 22:16:02 -04003055out:
Al Viroed782b52016-03-09 19:52:39 -05003056 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003057 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003058}
Al Virob5ae6b12014-10-12 22:16:02 -04003059EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003060
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003061static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01003062{
3063 *buflen -= namelen;
3064 if (*buflen < 0)
3065 return -ENAMETOOLONG;
3066 *buffer -= namelen;
3067 memcpy(*buffer, str, namelen);
3068 return 0;
3069}
3070
Waiman Long232d2d62013-09-09 12:18:13 -04003071/**
3072 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003073 * @buffer: buffer pointer
3074 * @buflen: allocated length of the buffer
3075 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003076 *
3077 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3078 * make sure that either the old or the new name pointer and length are
3079 * fetched. However, there may be mismatch between length and pointer.
3080 * The length cannot be trusted, we need to copy it byte-by-byte until
3081 * the length is reached or a null byte is found. It also prepends "/" at
3082 * the beginning of the name. The sequence number check at the caller will
3083 * retry it again when a d_move() does happen. So any garbage in the buffer
3084 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003085 *
3086 * Data dependency barrier is needed to make sure that we see that terminating
3087 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003088 */
Al Viro9aba36d2016-07-20 22:28:45 -04003089static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003090{
Waiman Long232d2d62013-09-09 12:18:13 -04003091 const char *dname = ACCESS_ONCE(name->name);
3092 u32 dlen = ACCESS_ONCE(name->len);
3093 char *p;
3094
Al Viro6d13f692014-09-29 14:46:30 -04003095 smp_read_barrier_depends();
3096
Waiman Long232d2d62013-09-09 12:18:13 -04003097 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003098 if (*buflen < 0)
3099 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003100 p = *buffer -= dlen + 1;
3101 *p++ = '/';
3102 while (dlen--) {
3103 char c = *dname++;
3104 if (!c)
3105 break;
3106 *p++ = c;
3107 }
3108 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003109}
3110
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003112 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003113 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003114 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003115 * @buffer: pointer to the end of the buffer
3116 * @buflen: pointer to buffer length
3117 *
Waiman Long18129972013-09-12 10:55:35 -04003118 * The function will first try to write out the pathname without taking any
3119 * lock other than the RCU read lock to make sure that dentries won't go away.
3120 * It only checks the sequence number of the global rename_lock as any change
3121 * in the dentry's d_seq will be preceded by changes in the rename_lock
3122 * sequence number. If the sequence number had been changed, it will restart
3123 * the whole pathname back-tracing sequence again by taking the rename_lock.
3124 * In this case, there is no need to take the RCU read lock as the recursive
3125 * parent pointer references will keep the dentry chain alive as long as no
3126 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003127 */
Al Viro02125a82011-12-05 08:43:34 -05003128static int prepend_path(const struct path *path,
3129 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003130 char **buffer, int *buflen)
3131{
Al Viroede4ceb2013-11-13 07:45:40 -05003132 struct dentry *dentry;
3133 struct vfsmount *vfsmnt;
3134 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003135 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003136 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003137 char *bptr;
3138 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003139
Al Viro48f5ec22013-09-09 15:22:25 -04003140 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003141restart_mnt:
3142 read_seqbegin_or_lock(&mount_lock, &m_seq);
3143 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003144 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003145restart:
3146 bptr = *buffer;
3147 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003148 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003149 dentry = path->dentry;
3150 vfsmnt = path->mnt;
3151 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003152 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003153 while (dentry != root->dentry || vfsmnt != root->mnt) {
3154 struct dentry * parent;
3155
3156 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003157 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003158 /* Escaped? */
3159 if (dentry != vfsmnt->mnt_root) {
3160 bptr = *buffer;
3161 blen = *buflen;
3162 error = 3;
3163 break;
3164 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003165 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003166 if (mnt != parent) {
3167 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3168 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003169 vfsmnt = &mnt->mnt;
3170 continue;
3171 }
Waiman Long232d2d62013-09-09 12:18:13 -04003172 if (!error)
3173 error = is_mounted(vfsmnt) ? 1 : 2;
3174 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003175 }
3176 parent = dentry->d_parent;
3177 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003178 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003179 if (error)
3180 break;
3181
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003182 dentry = parent;
3183 }
Al Viro48f5ec22013-09-09 15:22:25 -04003184 if (!(seq & 1))
3185 rcu_read_unlock();
3186 if (need_seqretry(&rename_lock, seq)) {
3187 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003188 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003189 }
Al Viro48f5ec22013-09-09 15:22:25 -04003190 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003191
3192 if (!(m_seq & 1))
3193 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003194 if (need_seqretry(&mount_lock, m_seq)) {
3195 m_seq = 1;
3196 goto restart_mnt;
3197 }
3198 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003199
Waiman Long232d2d62013-09-09 12:18:13 -04003200 if (error >= 0 && bptr == *buffer) {
3201 if (--blen < 0)
3202 error = -ENAMETOOLONG;
3203 else
3204 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003205 }
Waiman Long232d2d62013-09-09 12:18:13 -04003206 *buffer = bptr;
3207 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003208 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003209}
3210
3211/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003212 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003213 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003214 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003215 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003216 * @buflen: buffer length
3217 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003218 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003220 * Returns a pointer into the buffer or an error code if the
3221 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003222 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003223 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003224 *
Al Viro02125a82011-12-05 08:43:34 -05003225 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226 */
Al Viro02125a82011-12-05 08:43:34 -05003227char *__d_path(const struct path *path,
3228 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003229 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003231 char *res = buf + buflen;
3232 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003234 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003235 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003236
Al Viro02125a82011-12-05 08:43:34 -05003237 if (error < 0)
3238 return ERR_PTR(error);
3239 if (error > 0)
3240 return NULL;
3241 return res;
3242}
3243
3244char *d_absolute_path(const struct path *path,
3245 char *buf, int buflen)
3246{
3247 struct path root = {};
3248 char *res = buf + buflen;
3249 int error;
3250
3251 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003252 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003253
3254 if (error > 1)
3255 error = -EINVAL;
3256 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003257 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003258 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259}
3260
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003261/*
3262 * same as __d_path but appends "(deleted)" for unlinked files.
3263 */
Al Viro02125a82011-12-05 08:43:34 -05003264static int path_with_deleted(const struct path *path,
3265 const struct path *root,
3266 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003267{
3268 prepend(buf, buflen, "\0", 1);
3269 if (d_unlinked(path->dentry)) {
3270 int error = prepend(buf, buflen, " (deleted)", 10);
3271 if (error)
3272 return error;
3273 }
3274
3275 return prepend_path(path, root, buf, buflen);
3276}
3277
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003278static int prepend_unreachable(char **buffer, int *buflen)
3279{
3280 return prepend(buffer, buflen, "(unreachable)", 13);
3281}
3282
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003283static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3284{
3285 unsigned seq;
3286
3287 do {
3288 seq = read_seqcount_begin(&fs->seq);
3289 *root = fs->root;
3290 } while (read_seqcount_retry(&fs->seq, seq));
3291}
3292
Jan Bluncka03a8a702008-02-14 19:38:32 -08003293/**
3294 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003295 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003296 * @buf: buffer to return value in
3297 * @buflen: buffer length
3298 *
3299 * Convert a dentry into an ASCII path name. If the entry has been deleted
3300 * the string " (deleted)" is appended. Note that this is ambiguous.
3301 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003302 * Returns a pointer into the buffer or an error code if the path was
3303 * too long. Note: Callers should use the returned pointer, not the passed
3304 * in buffer, to use the name! The implementation often starts at an offset
3305 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003306 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003307 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003308 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003309char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003311 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003312 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003313 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003315 /*
3316 * We have various synthetic filesystems that never get mounted. On
3317 * these filesystems dentries are never used for lookup purposes, and
3318 * thus don't need to be hashed. They also don't need a name until a
3319 * user wants to identify the object in /proc/pid/fd/. The little hack
3320 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003321 *
3322 * Some pseudo inodes are mountable. When they are mounted
3323 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3324 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003325 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003326 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3327 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003328 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003329
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003330 rcu_read_lock();
3331 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003332 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003333 rcu_read_unlock();
3334
Al Viro02125a82011-12-05 08:43:34 -05003335 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003336 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 return res;
3338}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003339EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340
3341/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003342 * Helper function for dentry_operations.d_dname() members
3343 */
3344char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3345 const char *fmt, ...)
3346{
3347 va_list args;
3348 char temp[64];
3349 int sz;
3350
3351 va_start(args, fmt);
3352 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3353 va_end(args);
3354
3355 if (sz > sizeof(temp) || sz > buflen)
3356 return ERR_PTR(-ENAMETOOLONG);
3357
3358 buffer += buflen - sz;
3359 return memcpy(buffer, temp, sz);
3360}
3361
Al Viro118b2302013-08-24 12:08:17 -04003362char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3363{
3364 char *end = buffer + buflen;
3365 /* these dentries are never renamed, so d_lock is not needed */
3366 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003367 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003368 prepend(&end, &buflen, "/", 1))
3369 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003370 return end;
Al Viro118b2302013-08-24 12:08:17 -04003371}
David Herrmann31bbe162014-01-03 14:09:47 +01003372EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003373
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003374/*
Ram Pai6092d042008-03-27 13:06:20 +01003375 * Write full pathname from the root of the filesystem into the buffer.
3376 */
Al Virof6500802014-01-26 12:37:55 -05003377static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003378{
Al Virof6500802014-01-26 12:37:55 -05003379 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003380 char *end, *retval;
3381 int len, seq = 0;
3382 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003383
Al Virof6500802014-01-26 12:37:55 -05003384 if (buflen < 2)
3385 goto Elong;
3386
Al Viro48f5ec22013-09-09 15:22:25 -04003387 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003388restart:
Al Virof6500802014-01-26 12:37:55 -05003389 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003390 end = buf + buflen;
3391 len = buflen;
3392 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003393 /* Get '/' right */
3394 retval = end-1;
3395 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003396 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003397 while (!IS_ROOT(dentry)) {
3398 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003399
Ram Pai6092d042008-03-27 13:06:20 +01003400 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003401 error = prepend_name(&end, &len, &dentry->d_name);
3402 if (error)
3403 break;
Ram Pai6092d042008-03-27 13:06:20 +01003404
3405 retval = end;
3406 dentry = parent;
3407 }
Al Viro48f5ec22013-09-09 15:22:25 -04003408 if (!(seq & 1))
3409 rcu_read_unlock();
3410 if (need_seqretry(&rename_lock, seq)) {
3411 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003412 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003413 }
3414 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003415 if (error)
3416 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003417 return retval;
3418Elong:
3419 return ERR_PTR(-ENAMETOOLONG);
3420}
Nick Pigginec2447c2011-01-07 17:49:29 +11003421
3422char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3423{
Waiman Long232d2d62013-09-09 12:18:13 -04003424 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003425}
3426EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003427
3428char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3429{
3430 char *p = NULL;
3431 char *retval;
3432
Al Viroc1031352010-06-06 22:31:14 -04003433 if (d_unlinked(dentry)) {
3434 p = buf + buflen;
3435 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3436 goto Elong;
3437 buflen++;
3438 }
3439 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003440 if (!IS_ERR(retval) && p)
3441 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003442 return retval;
3443Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003444 return ERR_PTR(-ENAMETOOLONG);
3445}
3446
Linus Torvalds8b19e342013-09-12 10:35:47 -07003447static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3448 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003449{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003450 unsigned seq;
3451
3452 do {
3453 seq = read_seqcount_begin(&fs->seq);
3454 *root = fs->root;
3455 *pwd = fs->pwd;
3456 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003457}
3458
Ram Pai6092d042008-03-27 13:06:20 +01003459/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 * NOTE! The user-level library version returns a
3461 * character pointer. The kernel system call just
3462 * returns the length of the buffer filled (which
3463 * includes the ending '\0' character), or a negative
3464 * error value. So libc would do something like
3465 *
3466 * char *getcwd(char * buf, size_t size)
3467 * {
3468 * int retval;
3469 *
3470 * retval = sys_getcwd(buf, size);
3471 * if (retval >= 0)
3472 * return buf;
3473 * errno = -retval;
3474 * return NULL;
3475 * }
3476 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003477SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478{
Linus Torvalds552ce542007-02-13 12:08:18 -08003479 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003480 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003481 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482
3483 if (!page)
3484 return -ENOMEM;
3485
Linus Torvalds8b19e342013-09-12 10:35:47 -07003486 rcu_read_lock();
3487 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488
Linus Torvalds552ce542007-02-13 12:08:18 -08003489 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003490 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003491 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003492 char *cwd = page + PATH_MAX;
3493 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003495 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003496 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003497 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003498
Al Viro02125a82011-12-05 08:43:34 -05003499 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003500 goto out;
3501
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003502 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003503 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003504 error = prepend_unreachable(&cwd, &buflen);
3505 if (error)
3506 goto out;
3507 }
3508
Linus Torvalds552ce542007-02-13 12:08:18 -08003509 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003510 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003511 if (len <= size) {
3512 error = len;
3513 if (copy_to_user(buf, cwd, len))
3514 error = -EFAULT;
3515 }
Nick Piggin949854d2011-01-07 17:49:37 +11003516 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003517 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
3520out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003521 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 return error;
3523}
3524
3525/*
3526 * Test whether new_dentry is a subdirectory of old_dentry.
3527 *
3528 * Trivially implemented using the dcache structure
3529 */
3530
3531/**
3532 * is_subdir - is new dentry a subdirectory of old_dentry
3533 * @new_dentry: new dentry
3534 * @old_dentry: old dentry
3535 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003536 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3537 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003538 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3539 */
3540
Yaowei Baia6e57872015-11-17 14:40:11 +08003541bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542{
Yaowei Baia6e57872015-11-17 14:40:11 +08003543 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003544 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003546 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003547 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003548
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003549 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003552 /*
3553 * Need rcu_readlock to protect against the d_parent trashing
3554 * due to d_move
3555 */
3556 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003557 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003558 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003559 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003560 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003561 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563
3564 return result;
3565}
3566
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003567static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003569 struct dentry *root = data;
3570 if (dentry != root) {
3571 if (d_unhashed(dentry) || !dentry->d_inode)
3572 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003574 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3575 dentry->d_flags |= DCACHE_GENOCIDE;
3576 dentry->d_lockref.count--;
3577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003579 return D_WALK_CONTINUE;
3580}
Nick Piggin58db63d2011-01-07 17:49:39 +11003581
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003582void d_genocide(struct dentry *parent)
3583{
3584 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585}
3586
Al Viro60545d02013-06-07 01:20:27 -04003587void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588{
Al Viro60545d02013-06-07 01:20:27 -04003589 inode_dec_link_count(inode);
3590 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003591 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003592 !d_unlinked(dentry));
3593 spin_lock(&dentry->d_parent->d_lock);
3594 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3595 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3596 (unsigned long long)inode->i_ino);
3597 spin_unlock(&dentry->d_lock);
3598 spin_unlock(&dentry->d_parent->d_lock);
3599 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600}
Al Viro60545d02013-06-07 01:20:27 -04003601EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602
3603static __initdata unsigned long dhash_entries;
3604static int __init set_dhash_entries(char *str)
3605{
3606 if (!str)
3607 return 0;
3608 dhash_entries = simple_strtoul(str, &str, 0);
3609 return 1;
3610}
3611__setup("dhash_entries=", set_dhash_entries);
3612
3613static void __init dcache_init_early(void)
3614{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003615 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616
3617 /* If hashes are distributed across NUMA nodes, defer
3618 * hash allocation until vmalloc space is available.
3619 */
3620 if (hashdist)
3621 return;
3622
3623 dentry_hashtable =
3624 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003625 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003626 dhash_entries,
3627 13,
3628 HASH_EARLY,
3629 &d_hash_shift,
3630 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003631 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003632 0);
3633
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003634 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003635 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636}
3637
Denis Cheng74bf17c2007-10-16 23:26:30 -07003638static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003640 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641
3642 /*
3643 * A constructor could be added for stable state like the lists,
3644 * but it is probably not worth it because of the cache nature
3645 * of the dcache.
3646 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003647 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003648 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
3650 /* Hash may have been set up in dcache_init_early */
3651 if (!hashdist)
3652 return;
3653
3654 dentry_hashtable =
3655 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003656 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 dhash_entries,
3658 13,
3659 0,
3660 &d_hash_shift,
3661 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003662 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 0);
3664
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003665 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003666 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667}
3668
3669/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003670struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003671EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673EXPORT_SYMBOL(d_genocide);
3674
Linus Torvalds1da177e2005-04-16 15:20:36 -07003675void __init vfs_caches_init_early(void)
3676{
3677 dcache_init_early();
3678 inode_init_early();
3679}
3680
Mel Gorman4248b0d2015-08-06 15:46:20 -07003681void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003684 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685
Denis Cheng74bf17c2007-10-16 23:26:30 -07003686 dcache_init();
3687 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003688 files_init();
3689 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003690 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691 bdev_cache_init();
3692 chrdev_init();
3693}